To a reviewer with three hours to spare, fake research is indistinguishable from good research. Unless you want to give me months to review a paper, I will not be able to detect fraud.

Instead I rely on the fact that the community will attempt to integrate these ideas after publication and arrive at the conclusion that the paper was garbage.

If you only have three hours, can't detect fraud, can't tell if the the paper is even real. Then what's the point of even doing the peer review? Again, peer review offered no value at all in this case.

Peer review can still inspect whether the evidence presented in a paper would—if not fabricated—truly be enough to support the paper's conclusion. Mistaken analysis of honestly collected data, improper experimental methods, flawed proofs, etc. can all show up at review time.

You can certainly come to that conclusion if you'd like. Arxiv exists. But "peer review isn't especially usrful" and "these entire fields are nonsense" are very different conclusions.