Nothing for my part, but there was an Azure "maintenance" on 4/16 where both the Win8 and Win10 buildbots were migrated to "newer" hardware (the first change to their physical machine for a really long time).  No word on why.  The VM machine class was unchanged, and the tests run on a local disk.  But perhaps there has been a performance or other impact (Spectre changes?)

That seems to line up in terms of timing, for example with Win10 3.x build 795 and Win8 3.x build 831 both running the AM of 4/16 with the first failures.

Oddly, I'm having trouble finding a reference to test_mmap in older Win10 3.x build logs, but on Win8 3.x builds it was definitely one of the slower tests before that point, although only in the 4-5 minute range.

So not sure.  Assuming it was the "upgrade" and given the persistence of the failure since, I suppose it's safer to assume this is a new normal for that platform - if skipping the particular test(s) would work, I say go for it at least for now.

-- David