To me, one of the more interesting aspects of that was that the ruby version optimised for reduced memory usage got a 2x speedup.

That makes sense in terms of cpu cache misses. Has anyone run any numbers to see how gcc -Ospace compares with -O2 on recent hardware?