Fwiw this, and more specifically the test in 6b853a9354dfe62df838b255b4af33b423e74642, triggers an assertion in threadpool. It somehow doesn't make the test report a failure but I think it can cause some spurious failures like in https://gitlab.winehq.org/rbernon/wine/-/jobs/56730#L1220 which doesn't seem related to the H264 decoder changes from the corresponding MR.