My Secret: I may or may not know the exact way to release the GIL on python 3.10+ I might also know how to recreate this minimally and easily without C extensions It's not a magic trick, or some weird off beat code that doesn't work. It functional, typed, code. What it really is I'll never actually reveal directly. Because I already did by releasing TokenGate , the abstraction itself is simple readable and traceable in that codebase. (Once you see it, you know why.) When I first unlocked the GIL I tested it and traced the exact requirements. You can do it with or without serialization (serializing makes it easier for obvious reasons once you know but the physics doesn't change). Threading can output ~45x operations across 32 workers on my system in varied normal task distributions.…