well, i would think a better comparison would be if a big pool like ghash.io gets hacked and the attacker iniates an succesful double-spend attack.
if this gets noticed fast the best way to deal with it would be a rollback
do you understand what a double spend attack is?
with 51% you cannot double spend anyone else's coins only the ones you control.
so lets say you hack the ghash.io and manage to control if for a few hours before someone notices (say everyone watching the pool was asleep, all miners absolutely everyone using the pool fell asleep)
so you point the pool to your mining address and mange to mine some 2000 or so bitcoins.
you then go ahead and transfer those coins to an exchange and buy up some LTC or some other coin with them. and then withdraw those LTC.
once you have your LTC you then use your massive 51% hashing power to fork the network so that your 2000 bitcoin deposit effectively never happened. <-- the double spend.
you have your 2000 btc and your bag of LTC...
meanwhile the miners and pool operators begin to wake up and take their hashing power off the pool.
this is what could happen in an absolute worst case scenario.
a few million dollars stolen from the BTC network.
given the size of the bitcoin network of some 8 billion dollars, I doubt that even that handsome some would be enough to trigger a roll-back which would destroy and entire 8 hours worth of transactions and shake confidence in the network.
we are talking economies of scale here... as was said by one of the previous posters.
btw the attack in mintpal was not a double spend it was a simple SQL injection attack.
but if the attacker was left to keep those 30% vericoins that were stolen then they could have used those coins for numerous attacks in the future.