@aliashraf: There's one variant of the "destructive attack" that could worry me:

If Bitcoin changed to an ASIC-unfriendly algorithm (like ProgPoW, where ASICs are only two times more efficient than CPU/GPUs, if I remember correctly), then the "destructive attacker" could simply buy standard computing equipment, if he has the funds to buy the necessary hashrate. The "PoW change" option then would become extremely difficult, as if we changed back to an ASIC-friendly algorithm, at the start there wouldn't be any ASICs available - so the attacker had still time to continue his attack with standard computing equipment.

So one could even argue that for Bitcoin it's better not to change to an ASIC-unfriendly algorithm, but to an algorithm suitable for differently constructed ASICs (Scrypt, for example?). I'm not totally sure about that, though.

@GaPR: I don't really understand your answer to my post ...