PoS proponents said it couldn't be done and kept insinuating that one would have to purchase up most of the currency to perform the attack but here is one example of a successful attack being committed on a real coin that was staking at 10 %:
https://bitcointalk.org/index.php?topic=897493.0 apexcoin attack successful with 0.07% of available supply (small, but the stake to destroy the coin was nothing- yes , you read that right, less than 1% stake!)
I'll tell you my theories anyway: the two PhDs who wrote that research have a clear nxt bias. Look at their hidden multibranch section: they say that they got 3/20 of 500 blocks with 10% stake. 3/20 of 500 is 75 blocks. They got 75 blocks! if people wait for 6 confirmations, I only need 6 blocks, not 75. So I can do with much less than 10% stake. Of course 10% stake means 10% of actively forging coins, which means 10% of 10% of supply. So, much less than 1% of supply to double spend. Of course that's still a few hundred BTCs for NXT so I'll start with a smaller coin.
I already bought a few coins... just wait until age is accumulated...
Other PoS coins are going to be targeted soon and with an end goal in attacking NxT eventually. A similar type of attack is claimed to be capable with around 6.6% of NxT(whether existing ownership, purchased, stolen , or borrowed by an exchange /bank)
The cost to attack a PoS coin is nothing because they can sell the PoS coin after the attack is successful but prior to the market having a chance to react to such attack. In theory an attacker could deposit the PoS coin onto an exchange, sell it for BTC, withdraw the BTC then double spend the deposit onto another exchange and repeat