Collusion in 'linear' voting loses the voter their preference in exchange for some reward, double-dipping is a simple defection.
GP is talking about personal force multiplication through collaboration, it's win-win so you have a genuine prisoner's dilemma. Being unable to verify cooperation doesn't mean it won't happen in other ways.
I don't know that this is actually bad just that a simple comparison to linear votes doesn't hold up.
[0] https://en.wikipedia.org/wiki/Vote_pairing#United_States_pre...
Unfortunately, some people do like lying, or at least they don't mind it as much, and if you create a system which rewards liars more than honest people, you shouldn't be surprised if the liars end up winning and you create more liars.
The quadratic maths tends towards making the gains from collusion very large, which means there is incentive not to defect even if detection is uncertain.
And bloc votes and ballots which aren't necessarily secret do represent some of the more plausible use cases for quadratic voting (corporate governance)
But a conversation like "I was thinking four points on A but I see why you're all in on B now, tell you what how about we go two for two on it" strikes me as a good outcome in a voting system. The fact that linear voting doesn't have an equivalent of it doesn't imply we should consider this 'collusion' rather than 'cooperation'.
https://en.wikipedia.org/wiki/Superrationality#Prisoner's_di...
Since you can't verify that collusion has worked, the best strategy is to not do it.