What researchers want is discoverability and credit. Some researchers are already famous and own their own distribution via arxiv or Twitter but for new researchers publishing a paper at a prestigious conference is a way to bootstrap a reputation.
If you really wanna get rid of large publishers what you really need to solve is discoverability and you can solve it with a combination of virtual mini conferences and callouts on Twitter and Hacker News.
I've gotten some of the best constructive feedback on my blog posts from Hacker News directly, I don't see how that's different from peer review. I don't find it very likely that someone would even be willing to give me feedback unless they themselves are already subject matter experts.
But really the impact of a piece of computer science research is directly correlated to how many people directly use the OSS inspired by the research. People in industry know this, what's left is is for tenure committees to also take software impact into account.