Looks like there can be a more significant difference after 100,000 repetitions: http://rubybenchmark.com/reports/13. In general, the program is pretty dumb and my math isn't that great. Accepting patches: http://github.com/sandro/bench_press