All committers have signed the CLA.

The new implementation could block processing of messages, so add another source for instability.
You might have been lucky not to to face any issues so far, but I suspect there are some concurrency issue in this new implementation. I think they are currently not manifesting because only sending is implemented and not receiving

When using a dedicated thread (which paho-mqtt does with loop_start()) you will end up with callbacks called on another thread then main, when handing back control to python / asyncio, you will need to make sure to properly handle this (otherwise the async code will fail)

So overall I would prefer a aiomqtt implementation, it should already do all that heavy lifting.

Any information on the issues you experienced with aiomqtt? Are you sure that it was related to aiomqtt and not paho-mqtt itself? In the long run (even more so when we plan to extend proxy to sending and receive) it might be better invested time to fix aiomqtt instead of re-implementing part of the async wrapper logic.