But again, the interesting thing is that you can buy up 51% of a PoS coin, and then sell off your coins so that you no longer have 51%, but your history of having once owned 51% makes it possible to attack the network at any time, for free.
I doubt it.
Doubt all you want, the math doesn't lie.
Of course I was somewhat wrong by stating "for free" -- there would be *some* computing resources involved (and thus electricity costs, etc.), but nothing close to what 51% attacking a PoW coin would require. Several orders of magnitude difference.
If you once had a 51% stake, you can build a better blockchain than the other 49% can, starting from the point where you owned 51%. You develop this blockchain in secret, after selling off your coins (and profiting from it); and then release your secret blockchain to the world, and nodes will pick it up because it carries more stake than the 49% blockchain. Now not only do you have your profit from the original sales of the coin, you have your 51% back (to the extent that it's worth anything). All coins would not have to be consolidated under one address; in fact, I believe that doing so would prevent the attack in most PoS implementations.
This is a fundamental problem with PoS, and is why Peercoin incorporates both PoW and centralized checkpointing. Sunny King is not a stupid guy, he knows that PoS has a big flaw or else he would have implemented pure PoS for Peercoin.
You're absolutely correct. The client will look for and accept the longest blockchain. Period. It doesn't matter how it became longest or who controls it. All that matters is that the attacking chain is longer than the proper chain.