PyPy's focus so far has been on the JIT. The interpreter part is mostly an equivalent of a brain-dead C implementation. Overall it does not matter for long runn

PyPy's focus so far has been on the JIT. The interpreter part is mostly an equivalent of a brain-dead C implementation. Overall it does not matter for long running programs (like web servers), but the warmup times are kind of horrible. That said, in a language like Python, the interpretation overhead matters much less, because bytecodes are typically very "fat" and involve say multiple dictionary lookups (which are way more costly than bytecode dispatch lookup). If I were to optimize the interpreter, I would start with specialized bytecodes, that avoids fatness of the bytecodes first.