It's possible that they use an inefficient algo to process orders. Then, even if the queue is only a few megs, if the algo run time takes longer and longer depending on how many orders are in the queue, that could explain why the lag takes place. They upgraded to a server with SSD's, so they might have been bottlenecked by IO. If a part of their order processing algo relies on disk IO (obviously it does, since it needs to transactional and talk to a database), it could be that things were backing up.
The other exchanges aren't getting the same amount of traffic, and I'm not sure if they are also getting the same amount of fake micro-transactions that were allegedly bogging down MtGox.
Either way, I'm curious about this too, but I think the only people who can answer these Q's are at MtGox, and they would not divulge this info.
What data are they storing? Orders what else? How can that ever amount to any significant amount of I/O bandwidth?
They would have to read and rewrite the entire data in the orderbook on each order to come even close to this amount. And it would make absolutely no sense to do it.