Yep and this test being shown is actually saying that about 5 sync workers acting on thousands of requests is faster then python async workers.

Theoretically it makes no sense. A Task manager executing tasks in parallel to IO instead of blocking on IO should be faster... So the problem must be in the implementation.