Fork and real multi-core

One interpreter boots, takes a memory snapshot, and forks 4 worker processes from the image. Each worker is an independent interpreter with its own GIL, so the same job runs on real CPU cores in parallel. On header-less hosting the first visit reloads once to unlock SharedArrayBuffer.

Spawning workers...
The job: def _fn(n): return sum(i*i for i in range(n)) over 4 tasks of n=2,000,000.