With two canonical forms, and two upload channels for each - "Fetch as Google" and sitemap upload - it would not be unlikely that four different Data Centers are involved. And, that's not even considering the possibly of load swapping if one or more get bogged down.
Theoretically, the Caffeine platform was to minimize any DC being overloaded. However, the demands of Panda for the reevaluation of huge amounts of data cannot not have put an unexpected strain on the DCs. IMO, until Panda is well behind us, all Google data will be in an elevated state of flux.