@matrix reminds me of the time I benchmarked simple linked list transversal in java and C and the java implementation blew C out of the water.

I suspect it was because malloc(3) is slow.