Thanks for providing (2). You did not provide that before. The attack you're suggesting doesn't work.
Difficulty adjustments require 2016 blocks and can only change the difficulty by a factor of 4 per cycle of 2016 blocks. Your DOS attack, if effective, would slow down block creation while in effect but it would not substantially change the computational power required to rewrite any of the existing chain.
Not so sure. People are greedy. I think the attack would work (although I guess we will not see such an attack in the near future).
For example: Upload a modified Bitcoin client which accepts a bounty of 500000 BTC per block starting from a specific block. Rent Amazon and Google hashing power and produce some 20 blocks faster than the Bitcoin community does. During this time, produce 500000 BTC per fresh block and pay a large number of bitcoin addresses some extra Bitcoins (as a kind of bribery...). Then reduce the bounty again to 100 BTC per block (or, different approach, 50 BTC) and distribute the information on what you just have done to many forums.
Now there is this interesting dilemma: If a user/miner adopts your software and the new block chain...he gets a share of the extra coins. If he does not...he will not get a share of the extra coins. Of course you will also get a big share of extra coins. I am not so sure how many users will stick with the old block chain and not be tempted by the extra share they might get ... if they switch. (Of course, getting the modified software to accept the modified blocks is not a trivial engineering task - but it can be done).