One of the reasons why scientists are currently dependent on the journal industry is because the journals control peer review. Unfortunately, instead of merely reviewing manuscripts for correctness, the journals link peer review to "impact," which is a prospective measure of importance (i.e. whether they are likely to get many citations). Every journal you submit your manuscript to will put it through a new round of peer review looking for both correctness and impact. Massive amounts of time are wasted by working scientists and journal editors attempting to prospectively "grade" manuscripts on how important they will be (along with checking the correctness of the manuscript). Upon rejection for either correctness or impact, authors resubmit their manuscript to a lower tier journal (+/- revision) and wait for the next round of reviews. Rinse/repeat. Eventually, assuming the paper is not awful, it will be published.
How could we do it differently?
One idea is to disentangle impact from correctness and eliminate the journals from peer review. Scientists already review each others manuscripts pro-bono, so here's a proposal: Scientists submit manuscripts to bioRxiv as "pre-prints" for a nominal fee (say $1000) and request peer review. An editor selects 3-4 potential reviewers and reviewers are paid a nominal fee (~$150) to perform the review. The money paid by authors covers the reviewer fees and editors salaries. After peer review, authors can decide whether they want to submit their articles for additional peer review and fancy publication in a journal.
Impact is thereafter judged only by how many citations a paper gets, not which glossy magazine it gets published in.