Oi! A value was not initialized, so it just contained some random number. For some reason beyond me, this never showed up before (the bug has been there for at least 6-12 months). And it never showed up in my Valgrind tests either. And you're the only person who mentioned this. Very very odd.
Great to know you found something! I wanted to make sure I had everything ticked off before reaching out to you with the issue (did the audio, video driver updates etc routines first!).
So would this also attributing to or affecting how the buffer/syncing was being calculated? It seemed that after the corrupted numbers appeared the likelihood of it then having the buffer "drain" for that output device grew considerably. Once it started, it seemed never to be able to recover (pull it back in the opposite direction) until it hit either end of the buffer and restarted.