I believe I have answered my own question.
I have 4 MEG appliances using the MWG from SITE2 as their proxy for their updates.
When I configure the 4 MEG appliances to use SITE1 MWG as their proxy, SITE2's stats become very stable and SITE1's stats become crazy like the graph above...
Now I understand passing the MEG updates through a proxy is not ideal, but why on earth would they be throwing off the performance stats like that? (Again, user experience is the same in both sites, notwithstanding if the MEG appliances fetch their updates through the site or not.