So I would expect this is related to the inbound queuing discussed on casual conversation, blahaj, and reddthat. Not much is written about the initial sync, but I would think it is a related problem.
My personal tl:dr, because I didn't read it all or dive into the technical details:
There is only one inbound queue per remote instance, the bigger instances fill the queue up leading to sync problems. This gets worse over large distances (I didn't really read into why). I guess there was an assumption of more smaller distributed instances than some of the larger ones that popped up. It sounds like a code scaling problem.