I redesigned the thread pools, moved some synchronization logic into inline assembly and (as "prayers are not a very effective concurrency primitive"[1]) sacrificed a virgin calendar[2] to the scheduling gods.

Result: The overengineered regex engine has entered ludicrous speed. Full text regular expression scanning at >150GB/s.

(This is not a perfect run, the best I've seen for this query was 161.5GB/s.)

[1]: https://lord.io/text-editing-hates-you-too/
[2]: https://de.aliexpress.com/item/1005010079762965.html