This benchmark isn't really useful as you pointed out. Microbenchmarks are always tricky, but check out the other two posts I just wrote here (about Martin Thompson and the benchmarks on GitHub) for hopefully more realistic benchmarks.