Okay, this attack is one I was unaware of. But this kind of attack should only be a problem if the participants are "slashed" for voting on the wrong chain (e.g. "Slasher 2.0" as presented by Buterin), but not if they are slashed when voting on more than one chain (like in the first Slasher version and also in Poelstra's example). In the latter, the attacker cannot create a history where another participant is double-voting and so is punished, so in this sense there is no ambiguity. The nodes should simply follow the attacker's chain because it should have more "chain trust" (or similar "score"), but in that situation he is the "legit" validator so I don't see the problem.
He edited his document to address Slasher:
You guys may think I'm a stubborn PoS (e.g. NXT) shill. But I may even reconsider my position on PoS - in fact, after I read Vlad's blog posts about the Casper history about one year ago, I already got pretty skeptic on it, or at least I thought it would lead to over-complicated (and thus more exploitable) algorithms. The reason I got more optimistic on it were several forum discussions where it seemed that all really dangerous attacks on PoS would depend on very unlikely assumptions (e.g. stakeholders accepting cheap bribes) because of "altruism-prime" being stronger than initially thought.
There is only one provably secure PoS whitepaper that has security proofs dealing with every aspect which those efforts you refer to lack.
Ouroboros states clearly the tradeoffs required to achieve that security.
The requirement that most of the stake has to remain online at all times, informs that PoS does not function in the real world without an oligarchy.
In any power-law distribution of wealth, 50% of the money supply is held by the spenders not savers (i.e. the speculators who have no interest whatsoever in participating in stake forging).