If those 20.000 files are not on 20.000 separate disks the performance gained is marginal over max one process per core because the 20.000 threads are mostly wa

If those 20.000 files are not on 20.000 separate disks the performance gained is marginal over max one process per core because the 20.000 threads are mostly waiting for input to be read from storage. I'd love to see a benchmark on the same machine and the same data that simply uses "find [...] -type f -exec grep -q ^23$ {} \; -print" - I'm pretty sure it would take much less time.