In the company I'm working for, we had to spent more engineer time on GIL workarounds (dealing with the extra complexity caused by multiprocessing, e.g. patchin

In the company I'm working for, we had to spent more engineer time on GIL workarounds (dealing with the extra complexity caused by multiprocessing, e.g. patching C++ libraries to put all their state into shared memory) than we needed for the Python 2 -> 3 migration. And we've only managed to parallelize less than half of our workload so far. Even if this will be a major breaking change to Python, it'll be worth it for us.