diff --git a/src/pkg/Make.deps b/src/pkg/Make.deps
index 44a4bdaa178..9df9db41653 100644
--- a/src/pkg/Make.deps
+++ b/src/pkg/Make.deps
@@ -1,5 +1,5 @@
 archive/tar.install: bufio.install bytes.install io.install os.install strconv.install
-base64.install: io.install os.install strconv.install
+base64.install: bytes.install io.install os.install strconv.install
 bignum.install: fmt.install
 bufio.install: io.install os.install utf8.install
 bytes.install: utf8.install
diff --git a/src/pkg/base64/base64.go b/src/pkg/base64/base64.go
index b680cf79fc1..3db0a83da93 100644
--- a/src/pkg/base64/base64.go
+++ b/src/pkg/base64/base64.go
@@ -6,6 +6,7 @@
 package base64
 
 import (
+	"bytes";
 	"io";
 	"os";
 	"strconv";
@@ -56,7 +57,7 @@ var URLEncoding = NewEncoding(encodeURL);
 
 // Encode encodes src using the encoding enc, writing
 // EncodedLen(len(input)) bytes to dst.
-// 
+//
 // The encoding pads the output to a multiple of 4 bytes,
 // so Encode is not appropriate for use on individual blocks
 // of a large data stream.  Use NewEncoder() instead.
@@ -66,6 +67,11 @@ func (enc *Encoding) Encode(src, dst []byte) {
 	}
 
 	for len(src) > 0 {
+		dst[0] = 0;
+		dst[1] = 0;
+		dst[2] = 0;
+		dst[3] = 0;
+
 		// Unpack 4x 6-bit source blocks into a 4 byte
 		// destination quantum
 		switch len(src) {
@@ -101,120 +107,76 @@ func (enc *Encoding) Encode(src, dst []byte) {
 	}
 }
 
-// encodeBlocker is a restricted FIFO for byte data that always
-// returns byte arrays whose lengths are some multiple of 3.
-type encodeBlocker struct {
-	// The overflow buffer contains data that should be returned
-	// before any data in nextbuf.
-	buffer [3]byte;
-	bufpos int;
-	nextbuf []byte;
-}
-
-// put appends the data contained in buf to the encode blocker's
-// buffer.  In general, you have to get everything out before you can
-// put another array.
-func (eb *encodeBlocker) put(buf []byte) {
-	if eb.nextbuf != nil {
-		panic("there is already a nextbuf");
-	}
-
-	// If we have anything in the overflow buffer, fill it up the
-	// rest of the way so we can return the overflow buffer.
-	bpos := 0;
-	if eb.bufpos != 0 {
-		for ; eb.bufpos < 3 && bpos < len(buf); eb.bufpos++ {
-			eb.buffer[eb.bufpos] = buf[bpos];
-			bpos++;
-		}
-	}
-
-	if bpos < len(buf) {
-		eb.nextbuf = buf[bpos:len(buf)];
-	}
-}
-
-// get retrieves an input quantum aligned byte array from the encode
-// blocker.
-func (eb *encodeBlocker) get() []byte {
-	// If there is data in the overflow buffer, return it first
-	if eb.bufpos > 0 {
-		if eb.bufpos < 3 {
-			// We don't have a full quantum
-			return nil;
-		}
-		eb.bufpos = 0;
-		return &eb.buffer;
-	}
-
-	// No overflow buffer, so return nextbuf.  However, it has to
-	// be quantum-aligned, so copy the tail of the data into the
-	// overflow buffer for next time.
-	end := len(eb.nextbuf)/3*3;
-	for i := end; i < len(eb.nextbuf); i++ {
-		eb.buffer[eb.bufpos] = eb.nextbuf[i];
-		eb.bufpos++;
-	}
-	b := eb.nextbuf[0:end];
-	eb.nextbuf = nil;
-	if end == 0 {
-		return nil;
-	}
-	return b;
-}
-
-// size returns the number of bytes remaining in the encode blocker's
-// buffer.
-func (eb *encodeBlocker) size() int {
-	return (eb.bufpos + len(eb.nextbuf))/3*3;
-}
-
 type encoder struct {
-	w io.Writer;
-	enc *Encoding;
 	err os.Error;
-	eb encodeBlocker;
+	enc *Encoding;
+	w io.Writer;
+	buf [3]byte;		// buffered data waiting to be encoded
+	nbuf int;			// number of bytes in buf
+	out [1024]byte;		// output buffer
 }
 
-func (e *encoder) Write(b []byte) (int, os.Error) {
+func (e *encoder) Write(p []byte) (n int, err os.Error) {
 	if e.err != nil {
 		return 0, e.err;
 	}
 
-	e.eb.put(b);
-
-	output := make([]byte, e.eb.size()/3*4);
-	opos := 0;
-
-	for {
-		block := e.eb.get();
-		if block == nil {
-			break;
+	// Leading fringe.
+	if e.nbuf > 0 {
+		var i int;
+		for i = 0; i < len(p) && e.nbuf < 3; i++ {
+			e.buf[e.nbuf] = p[i];
+			e.nbuf++;
 		}
-		e.enc.Encode(block, output[opos:len(output)]);
-		opos += len(block)/3*4;
+		n += i;
+		p = p[i:len(p)];
+		if e.nbuf < 3 {
+			return;
+		}
+		e.enc.Encode(&e.buf, &e.out);
+		var _ int;
+		if _, e.err = e.w.Write(e.out[0:4]); e.err != nil {
+			return n, e.err;
+		}
+		e.nbuf = 0;
 	}
 
-	n, err := e.w.Write(output);
-	if err != nil {
-		e.err = io.ErrShortWrite;
-		return n/4*3, e.err;
+	// Large interior chunks.
+	for len(p) > 3 {
+		nn := len(e.out)/4 * 3;
+		if nn > len(p) {
+			nn = len(p);
+		}
+		nn -= nn % 3;
+		if nn > 0 {
+			e.enc.Encode(p[0:nn], &e.out);
+			var _ int;
+			if _, e.err = e.w.Write(e.out[0:nn/3*4]); e.err != nil {
+				return n, e.err;
+			}
+		}
+		n += nn;
+		p = p[nn:len(p)];
 	}
-	return len(b), nil;
+
+	// Trailing fringe.
+	for i := 0; i < len(p); i++ {
+		e.buf[i] = p[i];
+	}
+	e.nbuf = len(p);
+	n += len(p);
+	return;
 }
 
-// Close flushes any pending output from the encoder.  It is an error
-// to call Write after calling Close.
+// Close flushes any pending output from the encoder.
+// It is an error to call Write after calling Close.
 func (e *encoder) Close() os.Error {
 	// If there's anything left in the buffer, flush it out
-	if e.err == nil && e.eb.bufpos > 0 {
-		var output [4]byte;
-		e.enc.Encode(e.eb.buffer[0:e.eb.bufpos], &output);
-		e.eb.bufpos = 0;
-		n, err := e.w.Write(&output);
-		if err != nil {
-			e.err = io.ErrShortWrite;
-		}
+	if e.err == nil && e.nbuf > 0 {
+		e.enc.Encode(e.buf[0:e.nbuf], &e.out);
+		e.nbuf = 0;
+		var _ int;
+		_, e.err = e.w.Write(e.out[0:4]);
 	}
 	return e.err;
 }
@@ -225,7 +187,7 @@ func (e *encoder) Close() os.Error {
 // writing, the caller must Close the returned encoder to flush any
 // partially written blocks.
 func NewEncoder(enc *Encoding, w io.Writer) io.WriteCloser {
-	return &encoder{w: w, enc: enc};
+	return &encoder{enc: enc, w: w};
 }
 
 // EncodedLen returns the length in bytes of the base64 encoding
@@ -254,7 +216,7 @@ func (enc *Encoding) decode(src, dst []byte) (n int, end bool, err os.Error) {
 		var dbuf [4]byte;
 		dlen := 4;
 
-dbufloop:
+	dbufloop:
 		for j := 0; j < 4; j++ {
 			in := src[i*4+j];
 			if in == '=' && j >= 2 && i == len(src)/4 - 1 {
@@ -305,193 +267,68 @@ func (enc *Encoding) Decode(src, dst []byte) (n int, err os.Error) {
 	return;
 }
 
-// quantumReader wraps a regular reader and ensures that each read
-// will return a slice whose length is a multiple of 4-bytes.
-type quantumReader struct {
-	r io.Reader;
-	buf [4]byte;
-	buflen int;
-}
-
-func (q *quantumReader) Read(p []byte) (int, os.Error) {
-	// Copy buffered data into the output
-	for i := 0; i < q.buflen; i++ {
-		p[i] = q.buf[i];
-	}
-
-	// Read more data into the output
-	n, err := q.r.Read(p[q.buflen:len(p)]);
-
-	// Buffer tail data that does not fit into the quanta
-	end := (q.buflen+n)/4*4;
-	for i := end; i < q.buflen+n; i++ {
-		q.buf[i-end] = p[i];
-	}
-
-	// Is EOF misaligned?
-	if err == os.EOF && q.buflen > 0 {
-		err = io.ErrUnexpectedEOF;
-	}
-
-	return end, err;
-}
-
-// decodeBlocker takes a sequence of arbitrary size output byte slices
-// and makes them available as a stream of byte slices whose lengths
-// are always a multiple of 3.
-type decodeBlocker struct {
-	output []byte;
-	noutput int;
-	overflow [3]byte;
-	overflowstart int;
-}
-
-// flush flushes as much data from the overflow buffer as possible in
-// to the current output buffer, reseting the output buffer to nil if
-// it fills it up.  It returns the number of bytes written to the
-// output buffer.
-func (db *decodeBlocker) flush() int {
-	// Copy overflow into the beginning of this buffer
-	i := 0;
-	for ; i < len(db.output) && db.overflowstart < 3; i++ {
-		db.output[i] = db.overflow[db.overflowstart];
-		db.overflowstart++;
-	}
-	if i == len(db.output) {
-		db.output = nil;
-	} else {
-		db.output = db.output[i:len(db.output)];
-	}
-	return i;
-}
-
-// use begins using a new output buffer.  Any data that did not fit in
-// the previous output buffer will be placed at the beginning of this
-// buffer.
-func (db *decodeBlocker) use(buf []byte) {
-	db.output = buf;
-	db.noutput = 0;
-	// Copy left-over overflow from the previous buffer into this
-	// buffer
-	db.noutput += db.flush();
-}
-
-// checkout retrieve the next slice to fill with data.  The length of
-// the returned slice will always be a multiple of 3.  It returns nil
-// if there is no more buffer space.
-func (db *decodeBlocker) checkout() []byte {
-	// If we can use the output buffer, do so
-	if len(db.output) >= 3 {
-		end := len(db.output)/3*3;
-		return db.output[0:end];
-	} else if db.overflowstart == 3 {
-		// Fill the overflow buffer
-		db.overflowstart = 0;
-		return &db.overflow;
-	}
-	// We're out of space
-	return nil;
-}
-
-// checking indicates that we're done with the checked-out slice and
-// that we wrote count bytes to it.
-func (db *decodeBlocker) checkin(count int) {
-	if db.overflowstart == 3 {
-		// Wrote to the output buffer
-		db.noutput += count;
-		db.output = db.output[count:len(db.output)];
-	} else {
-		// Wrote to the overflow buffer.  Flush what we can to
-		// the output buffer.
-		n := db.flush();
-		if n > count {
-			n = count;
-		}
-		db.noutput += n;
-	}
-}
-
-// remaining returns the number of bytes remaining in the decode
-// blocker's buffer.  This will always be a multiple of 3.
-func (db *decodeBlocker) remaining() int {
-	return (len(db.output)+2)/3*3;
-}
-
-// outlen returns the number of bytes written to the output buffer.
-func (db *decodeBlocker) outlen() int {
-	return db.noutput;
-}
-
 type decoder struct {
-	r quantumReader;
-	enc *Encoding;
-	db decodeBlocker;
 	err os.Error;
-	// Have we definitely reached the end of the message?
-	end bool;
+	enc *Encoding;
+	r io.Reader;
+	end bool;		// saw end of message
+	buf [1024]byte;	// leftover input
+	nbuf int;
+	out []byte;		// leftover decoded output
+	outbuf [1024/4*3]byte;
 }
 
-func min(a int, b int) int {
-	if a < b {
-		return a;
-	}
-	return b;
-}
-
-func (d *decoder) Read(output []byte) (int, os.Error) {
+func (d *decoder) Read(p []byte) (n int, err os.Error) {
 	if d.err != nil {
 		return 0, d.err;
 	}
 
-	d.db.use(output);
-
-	var inbuf [512]byte;
-
-	// Read enough data to fill either our input buffer or our
-	// output buffer.
-	maxin := min(d.db.remaining()/3*4, len(inbuf));
-	n, err := d.r.Read(inbuf[0:maxin]);
-
-	// Decode into output buffer.
-	ipos := 0;
-	for ipos < n {
-		outbuf := d.db.checkout();
-		if outbuf == nil {
-			// Out of output buffer space
-			break;
-		}
-
-		inlen := min(len(outbuf)/3*4, n - ipos);
-		if d.end {
-			// We've seen end-of-message padding, but
-			// there's more data.  The RFC says this is an
-			// error.
-			// XXX Should shift character count
-			d.err = CorruptInputError(0);
-			break;
-		}
-		count := 0;
-		count, d.end, d.err = d.enc.decode(inbuf[ipos:ipos+inlen], outbuf);
-		d.db.checkin(count);
-		if d.err != nil {
-			// XXX Should shift character count
-			break;
-		}
-		ipos += inlen;
+	// Use leftover decoded output from last read.
+	if len(d.out) > 0 {
+		n = bytes.Copy(p, d.out);
+		d.out = d.out[n:len(d.out)];
+		return n, nil;
 	}
 
-	if err != nil && d.err == nil {
+	// Read a chunk.
+	nn := len(p)/3*4;
+	if nn < 4 {
+		nn = 4;
+	}
+	if nn > len(d.buf) {
+		nn = len(d.buf);
+	}
+	nn, d.err = io.ReadAtLeast(d.r, d.buf[d.nbuf:nn], 4-d.nbuf);
+	d.nbuf += nn;
+	if d.nbuf < 4 {
+		return 0, d.err;
+	}
+
+	// Decode chunk into p, or d.out and then p if p is too small.
+	nr := d.nbuf/4 * 4;
+	nw := d.nbuf/4 * 3;
+	if nw > len(p) {
+		nw, d.end, d.err = d.enc.decode(d.buf[0:nr], &d.outbuf);
+		d.out = d.outbuf[0:nw];
+		n = bytes.Copy(p, d.out);
+		d.out = d.out[n:len(d.out)];
+	} else {
+		n, d.end, d.err = d.enc.decode(d.buf[0:nr], p);
+	}
+	d.nbuf -= nr;
+	for i := 0; i < d.nbuf; i++ {
+		d.buf[i] = d.buf[i+nr];
+	}
+
+	if d.err == nil {
 		d.err = err;
 	}
-
-	return d.db.outlen(), d.err;
+	return n, d.err;
 }
 
 // NewDecoder constructs a new base64 stream decoder.
 func NewDecoder(enc *Encoding, r io.Reader) io.Reader {
-	return &decoder{r: quantumReader{r:r},
-			enc: enc,
-			db: decodeBlocker{overflowstart: 3}};
+	return &decoder{enc: enc, r: r};
 }
 
 // DecodeLen returns the maximum length in bytes of the decoded data
diff --git a/src/pkg/base64/base64_test.go b/src/pkg/base64/base64_test.go
index 1071200a0df..d11d99a8819 100644
--- a/src/pkg/base64/base64_test.go
+++ b/src/pkg/base64/base64_test.go
@@ -6,6 +6,7 @@ package base64
 
 import (
 	"base64";
+	"bytes";
 	"io";
 	"os";
 	"reflect";
@@ -167,3 +168,36 @@ func TestDecodeCorrupt(t *testing.T) {
 		}
 	}
 }
+
+func TestBig(t *testing.T) {
+	n := 3*1000+1;
+	raw := make([]byte, n);
+	const alpha = "0123456789abcdefghijklmnopqrstuvwxyzABCDEFGHIJKLMNOPQRSTUVWXYZ";
+	for i := 0; i < n; i++ {
+		raw[i] = alpha[i%len(alpha)];
+	}
+	encoded := new(io.ByteBuffer);
+	w := NewEncoder(StdEncoding, encoded);
+	nn, err := w.Write(raw);
+	if nn != n || err != nil {
+		t.Fatalf("Encoder.Write(raw) = %d, %v want %d, nil", nn, err, n);
+	}
+	err = w.Close();
+	if err != nil {
+		t.Fatalf("Encoder.Close() = %v want nil", err);
+	}
+	decoded, err := io.ReadAll(NewDecoder(StdEncoding, encoded));
+	if err != nil {
+		t.Fatalf("io.ReadAll(NewDecoder(...)): %v", err);
+	}
+	
+	if !bytes.Equal(raw, decoded) {
+		var i int;
+		for i = 0; i < len(decoded) && i < len(raw); i++ {
+			if decoded[i] != raw[i] {
+				break;
+			}
+		}
+		t.Errorf("Decode(Encode(%d-byte string)) failed at offset %d", n, i);
+	}
+}