I think the DMUS_E_ALREADY_SENT failures should not happen anymore after d0c3a0e03d557498195a70693003deae5b3e4588 (or rather much less frequently as it's always timing dependent).
Right, forgot about that and just looked at the top of MR where I saw those failures.\ So those can be dropped then.
Regarding the timing test I'm not completely sure making it flaky is the best option. Maybe we should just get rid of the test entirely if there's too much variation.
You're not sure and I'm deferring the decision to you.\ So going with flaky is the simplest workaround for now to stop that noise. And then later on we can remove the whole test if it isn't that useful.