Turns out the optimal value is 8 on cortex-A9 systems (pandaboard)

benchmark                     old ns/op    new ns/op    delta
BenchmarkAppend                     907          908   +0.11%
BenchmarkAppend1Byte                101          101   +0.00%
BenchmarkAppend4Bytes               116          116   +0.00%
BenchmarkAppend8Bytes               139          138   -0.72%
BenchmarkAppend16Bytes              185          158  -14.59%
BenchmarkAppend32Bytes              131          131   +0.00%
BenchmarkAppendStr1Byte              72           72   +0.00%
BenchmarkAppendStr4Bytes             93           93   -0.21%
BenchmarkAppendStr8Bytes            116          116   +0.00%
BenchmarkAppendStr16Bytes           161          125  -22.36%
BenchmarkAppendStr32Bytes           102          102   +0.00%
BenchmarkAppendSpecialCase          613          613   +0.00%

R=golang-dev, r
CC=golang-dev
https://golang.org/cl/8863045