Paper X, on the ArXiv, keeps getting rejected. Years later, Paper Y comes up and makes progress on X, or does something closely related to X. Y may or may not cite X. Y gets published. Now X cannot get published because the referees do not see what the contribution of X is, given that Y has been published and that in light of Y X is not new.
The solution in my opinion, following a series of earlier posts the last one of which is this, is to move the emphasis away from publication towards ArXiv appearance. Citations should refer to the first available version, often the ArXiv one. Journals/conferences can still exist in this new paradigm: their main important job would be to assign badges to ArXiv papers.
Obviously, this plan does not work for the entities behind current journals/conferences. So they enforce the status quo, and in the most degrading way: by forcing authors to fish out, maintain, and format useless references.