this was exhausting file descriptors with enough concurrent reads, from what I can tell. Either way, spawning a new OS thread per read is not the way. Because this is so critical, I added unit-tests for that behavior.
Settings
status