Tests re-run; in 800 instances we got a single metadata failure. After discussing the status with all in a teleconference, we decided to approve the changes so far:
* no regressions were detected;
* one failure in 800, on a extremely stressed UEC is *way better* than ~5 failures in 100.
* we need this fix published for stability.
So I am marking this verification-done; Eucalyptus is looking at the error, and will eventually propose an addendum to the patch, which we will integrate in another SRU.