This is a better explanation: https://web.archive.org/web/20180403212547/https://caml.inria.fr/pub/ml-archives/caml-list/2004/07/e86a25aa6c6a6a7d08dd7eb50cfd5d5

This is a better explanation: https://web.archive.org/web/20180403212547/https://caml.inria.fr/pub/ml-archives/caml-list/2004/07/e86a25aa6c6a6a7d08dd7eb50cfd5d52.fr.html https://web.archive.org/web/20180403212547/https://caml.inri... You'd be better off measuring actual performance on real world tasks before inferring anything. Modern CPUs overlap execution of instructions very aggressively.