I too am seeing datasets going into a nonconformant state. My errors are only on those datasets where I do a cascasding mirror (primary -> 1st mirror -> 2nd mirror). The second relationship seems to get lost on random volumes. I can see this by checking the snapmirror status on the 1st mirror array and finding only one relationship (there should be two). Easily corrected by going to the 2nd mirror controller and run "snapmirror resync" against the volume(s) in question. I just end up doing this for 5-10 volumes per day. And if I don't do it the other volumes in the dataset mirror successfully but those with errors are just skipped. I see this on both Oracle and SQL volumes (only ones doing cascading mirrors). My cascading SnapVault (primary -> mirror -> vault) are running happily.
I do have a case open with Support but since it is a low priority no one has actively worked it yet.