Is there some reason we can't just revert #18462? I'm still seeing slowdown in tests due to that.

@Andy-MS I'd rather not, since even if it added 20 seconds to normal test runtime (which is not uniformly observed - for @sandersn and I runtime has improved) - it significantly improves the runtime of jake runtests-parallel test=rwc - by many minutes - and it is not worth maintaining two parallel runners just because one is sometimes 5% better on some machines sometimes.

RWC is something that should be running on a machine whose time doesn't matter, but the performance and reliability of the regular test suite is something that directly effects my ability to get work done. See also #18695.

@Andy-MS If I, as a dev, make decisions based on RWC output (and I do), then it is in my best interest to see those results completed as quickly as possible (wherever they are run), so I may iterate on those choices quickly. I do actually try to run them on a separate machine (so as to not occupy my own); but I still want them to complete quickly there. ;)

And again - on many machines the new harness is faster; if on yours it is not, I am still interested in tracking down the reasons why; as the new parallel harness should have overall better resource utilization and lower overhead.