Thank you! I think I found the root cause – it’s a good old deadlock situation which halts all handling of the data, eventually resulting in increasing queue.
I actually managed to reproduce the issue, although not consistently (must be a timing thing)
I now pushed a fix, and it should be brewing here