Calls to io_uring_submit() with iouring do a user/kernel context switch, which is computationally expensive. When in external execution mode with iouring datapath, allow the app to do all calls to io_uring_submit, and have don't have MsQuic call io_uring_submit.
First steps would be seeing the current performance costs of having MsQuic call io_uring_submit and confirming that there's performance gains to be had.