With 33+% of hashing power I can break Bitcoin, this is what the paper says. https://people.eecs.berkeley.edu/~luca/cs174/byzantine.pdf talks about 33% too. 51% works only in unrealistic assumptions. I can imagine a case where an attacker with 90% of hashing power won't break the system. I hope you got my point, I'm not going to extend it because it's an off-topic.