Regarding the latter thing, there would be some edge cases and possibly implementation issues to work out (but it might be easy, I'm just not sure). I do think it would be an improvement over the status quo, but how much of a difference it would make in practice I don't know. Without getting more downvotes deployed somehow as well, probably not much.
Regarding how to do anonymous voting, yes likely some sort of zero knowledge rollup type thing as you suggested but with proper auditing should still be secure. I don't see us being resourced to undertake that sort of development at the moment but with 5 million USD in DAO and growing, perhaps we'll get there.
RE: Proof of Brain Theory & Further Optimization