Even today's smartphones aren't "simple" terminals (and probably
much more powerful than the mainframes he envisioned). The closest to that model is SSH and Onlive gaming. The trend is for webapps to do as much processing on the client as possible, and use the backend mainly for data. So I think he was wrong (and the article is also wrong in saying he was vindicated).
People have constantly predicted a mainframe model; but cheaper, simpler and more convenient processing devices keep seeping computing power towards the user. That is: processing power migrates to the edge of the network. But I don't think incorrect predictions of the far future should in any way be held against McCarthy.
AI tangent: my thoughts on a "hivemind" vs "individuals" (mainframe vs. distributed) is similar to capitalism vs. a planned economy - individuals can collect data from their own viewpoint; form their own hypotheses; and act on them locally. Of course, theoretically, a hivemind with the same processing power could do the same with waldos; however, I think it would be difficult to resist consolidating hypotheses, and concentrating on the best ones (as large companies kill off small projects that later disrupt them). Whereas, individuals will do things their own way - and some of them will come up with something better (like ants searching for sugar; or startups seeking disruption). Most of them end up dead of course. There's also practical concerns favouring distribution: network latency/bandwidth; power consumption and redundancy (not all eggs in one basket).