FWIW, there is no inherent reason io_uring should be slower. Both implementations look pretty basic to me (checked method_*.c files). Doing thread pool io is much easier than using io_uring properly with all features so it might be possible that it will be better in the future.
You can compare worker_pool/io_uring/aio etc. using `fio` to see how different io approaches perform on the same hardware.
You can compare worker_pool/io_uring/aio etc. using `fio` to see how different io approaches perform on the same hardware.