IIRC (and I remember it quite vaguely from some random blog post on the Internet) the Lua implementation uses some trick that greatly speeds up setting up and tearing down activation frames (and has also something to do with the ability to easily cross-call into C back to Lua again?), but it hinges on that argument-passing/result-returning semantics being "use nils for missing stuff, throw away the extras" so e.g. Python can't use it to speed up its implementation of function calls.

Unfortunately, I can't find that blog post so take my words with a grain of salt.