Since I didn't see any open issues about it and jimmo asked, I though I'd open this long-standing issue:

If you cancel a coroutine, the exception won't be raised before the coroutine is scheduled next. This is because it just stays in waitq until it would be executed regularly.
So a coroutine yielding an await uasyncio.sleep(10) will only receive the cancel exception after those 10 seconds have passed.

This results in very late exception handling inside the coroutine but also in canceled coroutines being "stuck" in waitq so the user code has to be careful not to cause a queue overflow if he uses cancellations often.

The fix however requires more than a change to uasyncio, it requires changes to utimeq too so a coroutine can be removed from the queue.
As I understand it Paul has that fixed on his fork.

@peterhinch implemented a high level workaround on his fast_io implementation for making coroutines cancel quickly by using an additional queue but even with that implementation coroutines stay in the utimeq until they are scheduled regularly.