Self-publishing is not without its problems.

Peer review, while not perfect, guarantees that the text has been reviewed by quality researchers. Journals have an interest in filtering bad publications, if only to keep their reputation.

In contrast, a self-published paper has no guarantee of quality, simply because everyone can do it. An author's reputation is often a proxy to assess a publication's quality, but this is clearly bad for new researchers.

I think there are good alternatives - my last publication costed around $300, which were used to pay for the conference where I presented it.

Ah, so we simply create a protocol that transparantly breaks down the peer review process into a rainbow of gradients thereby allowing the desired process of review by what you call quality reseachers. Such protocol could allow a review-for-free mechanism that could be time consuming and it could build a fund for the review process. Arguably the review process needs not to be finished at any time. It is probably big fun to revisit the process in the licht of later findings. Lastly but not least the protocol could help establish a name for reviewing scientists who just happen to do a lot of reading. A (large) bit like github does for programmers.

Self-publications and 'gray literature' are generally deemed worthless and often even considered to contribute a large amount of negative value to a CV. To put things simple, you'll loose your job.

Not if I want to keep my job come review time.