Is the kind of attack that you described can be successfully implemented without holding 51% of the coins?
I don't know if any math analysis has been done to work out at what % such an issue/attack would start to wreak havoc but as stated if all parties "acted selfishly" (which is what is assumed in Bitcoin) then there is no reason to think that even small % holders wouldn't create multiple blocks (one for each possible fork) so they would all end up colluding in what would then amount to most likely an even bigger problem than >50% (as the forks would never end).
Also note that to start a fork you only need to withhold your "better block" just long enough after an inferior one has been minted (clearly the more minting power you have the longer you can make such a fork before it is "published" but even a fork of 1 block could start such a chain reaction and that could be done unintentionally even).
It should be noted that "automatic checkpoint" mechanisms can limit the length of any particular fork but you are still potentially just going to start up a bunch of new forks after such a checkpoint has occurred (and once faith in such a blockchain has been dealt this kind of a blow I couldn't see the coin really having a future).
Basically POS only works if the coin is controlled by a cartel (which won't work against itself in creating and continually extending such forks) or if the (vast?) majority of players do not participate in such NAS behaviour (i.e. they only use the standard software that won't behave that way).
Those that argue "but we haven't seen this happen on XYZ" simply aren't admitting the most likely reason why that would be the case (i.e. because XYZ is actually under then control of a cartel and of course this is almost impossible to either prove or disprove). Also if an alt-coin that isn't worth very much is using POS then it is far less likely to be subject to such an issue as there simply isn't the financial incentive to care (and a smaller community is perhaps more likely to have more honest players).
Of course the standard software would have to be modified to take the NAS approach (but presumably that wouldn't cost much to do if there was potentially a lot of money to be made from wreaking such havoc).
IMO there are other ways that energy consumption could be dramatically reduced that don't carry such a risk (and I'll be elaborating more upon that later this year most likely).