Too bad I'm coming late to the party, but those interested by this post might be interested by the test implementations that I made a few months ago to bench various design parameters of interpreters (bytecode size, gCC computed goto extension...) in high level (I.e boost::variant<>) C++. I don't have the numbers at hand but I remember it was very CPU dependant. You can try for yourself, it is only a few self contained small programs : https://github.com/scientific-coder/Computer-Languages/tree/...