Whenever someone claims a proof (or disproof) of a big conjecture, many people leap to the question of whether the proof is correct. The problem then is it that it takes an enormous amount of work to confirm that a proof is correct. Even a clear mistake in a proof could be reparable. Moreover, attempted proofs have inferences that amount to gaps of different sizes. Even in a naive attempt, it can take a lot of work to decide which gaps are so big that the proof has to be called incomplete.
There is a much simpler standard that experts use in practice: "As I start to read this paper, am I learning from it?" You would expect a proof of a big conjecture to have very interesting lemmas, and otherwise to teach you new things along the way. This is not always obvious either; there have been a few grievous misunderstandings in which initial readers dismissed a great paper. Even so, it's a somewhat reliable standard, and it's the most that authors can expect.
When Perelman posted the first of his three papers on geometrization, experts in differential geometry quickly embraced it as exciting and teachable, before they had even checked half of that paper or seen the other two papers. From the beginning, this was very different from most claimed proofs of the Poincare conjecture, even most of the noble failures. The great ideas in these papers were more important than the fact that they had a lot of gaps (by common standards) and even some inessential mistakes (or so I was told).
I know for a fact that experts sometimes do study weird-looking claims of big results, in the arXiv and elsewhere. They have little incentive to broadcast their attention to it if they think that it's shoddy work, but sometimes they try to be fair. For starters, the math arXiv has moderators, and they often take a look. I think that usually (not quite always), several people have looked long enough to decide that they aren't getting anything out of the paper. But hey, there could always be a diamond in the rough, or even a diamond in the garbage.