I've changed the approach to moving tests around so the tests that care about focus are first, and all of those tests should be run without focus stealing prevention being activated on Windows, at least sometimes. I don't think I can do better than that, but as long as Windows runs them correctly sometimes we can find out what the message sequence should be. That took a lot of time slowly sending jobs to the testbot to find intermittent Windows test failures that this introduced. I think I've noted in comments where I've seen each optional message, but unfortunately there's no way to keep this information up to date.
There are still a few corner cases where Wine is wrong. Most of these are cases where the tests already accounted for Wine activating the window when it shouldn't. WmRestore_3 is a case where Wine apparently never activates the Window. SetActiveWindowSeq0 is a special case that probably needs to be handled separately, but I'm not sure how common it is.