Counter-point: we all know that "free" internet services are flaky. Eg many people prefer to pay for email, for reliability. This article introduces the perspective that OA journals are a bit like ad-supported email.

I don't really agree that traditional publishers lose all articles by default by this metric. I think there is some value in a reliable record of the journal itself, as more than the sum of its parts.

Still, my preferred solution for all of this is like JMLR -- very low-cost and open access, and it has reliability by virtue of association with a top university, and prestige by virtue of its editorial board (which becomes self-fulfilling).