In my case no, because I heavily use numpy, PyTorch, and other libraries that use C++ extensions and release the GIL during long operations. I easily use 24 CPU

In my case no, because I heavily use numpy, PyTorch, and other libraries that use C++ extensions and release the GIL during long operations. I easily use 24 CPU cores fully by submitting to one thread pool. If you use pure Python for long tasks, you're right that a ProcessPoolExecutor would be better.