What if we could look at authors, peer reviewers and papers as a graph of weighted edges to come up with a score that was independent on journals as a concept? And where there is an ontology for the semantics of edges (not only number of citations)?
Right, the perception of being associated with authoritative and knowledgeable publishers which follow certain formalities (in other words "the proof you as a scientist belong to the Club") has been an important part of academic career progression, even if in the last few years the reputation of peer-reviewed papers in general has taken a dive.
Frankly, they profit from: academics writing articles (funded by someone else); academics editing a journal deciding who should review articles (usually, but not always, for free); academics reviewing articles (always for free); and academics citing articles in subsequent ones.
The whole system is a house of cards, and a relatively immutable one.
Occasionally the editors -- inevitably based in Chennai -- spot a typo that the reviewers missed. Sometimes they cock things up massively, especially equations -- I had a big argument about a tickz-based scheme once. The formatting is done automatically and for me as a latex author I really think we could do that very easily. Colleagues who use word see the text transfer to something better as a major value add. The publicity the journal adds is a strong function of how "good" it is. Science and Nature effectively exist because they are Science and Nature.