Good reasoning, but I'd counter it with this: if the community of delegators as a whole agrees with the content that is being voted upon, as well as the amount that is earned, then isn't that still consistent with proof of brain? A possible exception would be if the centralized system goes against the direct interest of its delegators (i.e. upvoting content that the delegators do not want upvoted) - to my understanding, that would be counter to proof of brain, but then can be solved by undelegating.
(If I'm wrong, please do correct!)