The point is that Python object model prevents most optimizations, and thus makes the interpreter to executive code exactly as written. All inefficiencies become more obvious this way, while an optimizing compiler could have detected and eliminated many of them in a different language.
Yes, that’s a good point. But I think you can already see the difference in a lot of C++ codebases when you compile at -O0. Until recently, std::move() was a function call, and at -O0, this meant that calls which should be basically a memcpy() ended up having thousands of little function calls in it to std::move().
