I may be misunderstanding, but it doesn't look like we're solving the problem in the same way as Windows. Or, at least, you haven't confirmed that this patch is doing so.
Our current implementation returns aligned frames, whereas MFCalculateImageSize() returns an unaligned size. There are two possible fixes: either change our implementation to not use MFCalculateImageSize(), or change our implementation to return unaligned frames. It would surprise me if Windows returned aligned frames when MFCalculateImageSize() returns an unaligned size, and if so, the fix taken here would seem to be the wrong one.
Whether this is actually the case has not been confirmed, though.
Either way I feel like there probably should be some tests that are actually fixed in this series.