Just running machine code itself does not make a program magically faster. It‘s all about the amount of work the machine code is doing.
For example, if the JIT compiler realizes the program is adding two integers it could potentially replace the code with two MOVs and a single ADD. However, what about the error handling in the case of an overflow? Python switches to its internal BigInt representation in this case and cannot rely on architecture specific instructions alone once the result gets too large to fit into a register.
Modern programming languages are all about trading performance for convenience and that is what makes them slow — not because they are running an interpreter and not compiling to machine code.
For example, if the JIT compiler realizes the program is adding two integers it could potentially replace the code with two MOVs and a single ADD. However, what about the error handling in the case of an overflow? Python switches to its internal BigInt representation in this case and cannot rely on architecture specific instructions alone once the result gets too large to fit into a register.
Modern programming languages are all about trading performance for convenience and that is what makes them slow — not because they are running an interpreter and not compiling to machine code.