Python 3.14 subinterpreters cut a 3.6-second GIL bottleneck to 0.9 seconds

Python 3.14's new concurrent.interpreters module (PEP 734) delivers real parallelism: four subinterpreters ran a CPU-bound job in 0.9s versus 3.6s for four threads sharing the GIL. Each interpreter accepts only one concurrent call — a second raises InterpreterError immediately — and creating four subinterpreters (~0.077s) costs about twice as much as a fork-based process pool.

• 4 subinterpreters: 0.89–0.98s vs 3.51–3.70s for 4 threads
• A second concurrent call on one interpreter raises InterpreterError
• Creating 4 subinterpreters takes 0.076–0.078s vs 0.037s for fork
• Python 3.14 on Linux switched multiprocessing default from fork to forkserver