I had thought looking at the title that this was about a benchmark on a machine with one million hardware threads, which would have been interesting. Turns out

I had thought looking at the title that this was about a benchmark on a machine with one million hardware threads, which would have been interesting. Turns out to be a job of spawning loads of unprioritized identical threads that block, do negligible work and exit. It looks like time costs are going to be almost entirely dominated by setup and tear down, Which is very nearly useless for any situation in which you expect threads to exist for more than infinitesimal time. Setup and tear down are constants. They won't dominate the runtime of most programs.