This is part 1 of 2 for issue 4028

benchmark                       old ns/op    new ns/op    delta
BenchmarkReaderCopyOptimal          33495         9849  -70.60%
BenchmarkReaderCopyUnoptimal        70631        27041  -61.72%
BenchmarkReaderCopyOldImpl          51407        52970   +3.04%

Update #4028

R=dave, nigeltao, rsc, bradfitz, rogpeppe
CC=golang-dev
https://golang.org/cl/6548047