It's a costly business for the people doing the actual reviewing, ie the scholars who essentially work for free. Academic publishing is, I hope, the next industry to be reformed by open access web archives, although I'll be the first to admit that it's a long way away. So I think it's perfectly fine for you to be "fucked off".
Despite accepting basically everything that is valid science (typically a much lower bar than is required to publish in most journals), PLOS One has had a consistently increasing impact factor. In other words, the open access approach is getting some respect from scientists who produce the original material.
One of the best things about working in machine learning and natural language processing is that the best conference and journal archives are already open access ( books.nips.cc , usually ICML publishes the papers online, there's the ACL anthology, JMLR and JAIR are fully open access, etc).