51% attack is a term which signifies that an attacker has accumulated more computing power in his hands than all the other members of the network combined: a sort of controlling interest of generating power.
Consequences of 51% attack
51% attack vulnerability is a kind of cryptocuurency's sickness and a terminal one at that. When a new cryptocurrency isn't trusted it's network's power is poor so enough power for an attack can be accumulated easily. Of course there's no economical gain to be yielded at this stage but an attacker could, for example "kill off" a rising competitor.
Another possible way of application is an attack on already commercially successful currency for profit. However this case implies that the coin is commercially mined by network members who use their computing power for generation of cherished coins. In the case of a successful currency this means that the total network power is enormous as there's a massive number of people who are engaged in mining process. Thus a 51% attack becomes hard to conduct.
Therefore 51% attack is a tool of a natural selection of sorts that separates the coins with high potential from all others at early stages of thei existence. At the same time it may present a problem in the future when commcercial mining will come to an end. Once in every four years the reward for a block is halved. Now this reward equals 25 BTC per block and at the end of 2016 will drop to 12.5. Sooner or later this number will become lower than efforts spent on block generation. Lower people engaged in mining means lower combined network power means higher applicability of 51% attack.
51% attack and Bitcoin
This type of attack can also take place in Bitcoin network, if for example someone will buy an entire batch of BFL ASICs and commit them all to an attack. Although there will be no profit in it. Entire Bitcoin capitalization is around $5.5 billion at the moment. If someone starts withdrawing them at once their price will drop in an order of magnitude or two. Now one just has to compare this to the price of buying the equipment needed for such an attack. Most probably the attacker will be a manufacturer of mining equipment and not a buyer. What is more profitable in the end: to blow up a goldmine and get 10% of all gold in the process or continuously sell shovels to the prospectors for several years? Almost certainly it's the second.
There's also a dreamlike possibility of appearance of fundamentally new kinds of computing equipment, for example quantum computers. Although in this case Bitcoin price drop will be the least of our problems. In this scenario cryptographic resistance will be lost by most of current encryption systems including those used by governments and banks.