In the last decade, more and more researchers read the scientific literature on electronic copies or on printed copies, but not anymore directly from the printed journal. Nowadays, I think the most important role of journals is therefore only to offer a well-established editorial and refereeing platform.
Anyway, there is a sort of economic loophole. Let me explain. Research institutions pay both researchers and journals (to have online access and for printed copies) where researchers publish. On the other hand, journals offer a refereeing platform which is based on the unpaid work of referees, which are, again, researchers. Therefore, research institutions pay for all, and journals earn money for the research published. I could be a little naive here, but I think not far from truth.
Therefore, why are journals useful, apart from refereeing? For instance, imagine that arXiv, or another open-access archive will introduce refereeing and some sort of editorial filtering. In this scenario, will conventional journal publications still make sense?