Sorry, you're the second person to say this, so clearly I didn't phrase my comment very well. I meant that it's the interpreter which has undergone the most optimisation, or alternatively which has had the most brainpower dedicated to optimising it.

I don't mean that the resulting interpreter is the most performant interpreter in some given benchmark. Like I said in my other reply, I don't think that can possibly be a meaningful benchmark, given different languages have different grammars and semantics which have a huge effect on the possibility of writing a performant interpreter (see: Python).

I can see that you preempt this point in your comment, but, I mean, what on earth does "roughly comparable languages" mean? Comparable how? Even superficially similar languages can have semantic differences which have a deep impact on optimisability. (For instance Python's C ABI, invisible to most end users, which notoriously limits interpreter implementations. Or its bytecode dispatch. Or its attribute dynamism, precluding memoisation of lookups. The possibility of `eval` is an enormous one too, which also applies to JS.)