Paper on different attacks related to multibranching forging is published by Consensus Research https://github.com/ConsensusResearch/articles-papers/blob/master/multistrategy/multistrategy.pdf
TL/DR version and consequences:
- multibranch forging gives measurable possibility to earn more fees. I guess Nxt should not ignore it in long-term as the profitable activity will be implemented by somebody sooner or later
- there's no long-range attack against a blockchain V. Buterin described, only short-range. The short-range attack doesn't allow double-spending but gives multibranching forger possibility to earn more fees in singlebranch environment by producing few blocks in a row. However producing few blocks in a row could be an issue too (e.g. evil forger may postpones orders submissions etc) but not critical at the moment.
- not explicitly stated in the paper but easily derived, a long delay between blocks not only annoying but also a security problem as it's the moment for short-range attack could happens. Due to best chain measure chosen(cumulative difficulty) a long delay is a blockchain quality function minimum and multibranching forger can offer better solution in that time.
- we have formally defined nothing-at-stake attack(again, using Buterin's informal definition) and made initial simulations. We haven't included their results in paper as they are seems to be too raw, but I can reveal them here: N@S attack could happens only in short-range, e.g. for within 20 blocks for 10% stake, so with 30 blocks confirmation it's not possible to make the attack in our simulation. Also please note the attack has pretty unpredictable nature for attacker, so he can hardly enforce it, even in theory(in practice it's even harder to get it done properly). The correlation with stake size is still the open question, but it's nearly impossible to attack a proof-of-stake currency with "1% stake even" as stated by Buterin