I was too sleepy at the tail end of that marathon day yesterday when this debate heated up at the end with 4 against one sleepy one.
smooth et al, you forgot another key point. You argued the high difficulty would only cause the miners to win a block every day ~200 minutes instead 2 mins (assuming the more plausible 100X hashrate attack), but the miners will leave because they become unprofitable (as the block reward doesn't increase but the length of time they need to mine for the same reward increases by 100).
This may happen but a lot of miners are just on autopilot and won't stop. The hash rate will never literally go to zero. People mine on broken abandoned forks for months if not years. So maybe the 200 minutes becomes 500 or 1000 or whatever. It doesn't really matter because the longer the spiked block times, the faster the difficulty will drop once it starts dropping.
You earlier mentioned that the attacker can keep driving the difficulty up, and this is true, but it changes it from a one time attack to an ongoing attack. If you're going to do that, probably a regular 51% is at least as effective.
The in-progress Surae Noether difficulty algorithm (its somewhere on github) might also be more resistant to these types of attacks. It is designed to resist changes in hash rate that are "unnatural". I haven't looked carefully at this particular scenario though.
I know of one other way to improve resistance against this attack.