On the one hand, they're right about the "try me" effect. Many systems hype themselves by talking about the number of people who have ever used them, which is massively misleading -- for most of these things, huge numbers of people try them, say "eh", and leave. Second Life is certainly one of these: it's big, and reasonably popular, but isn't taking over the world to the degree that some of its proponents would like to claim.
And this article makes a good point that these "pure space" plays tend to never become really big. The analogy to LambdaMOO is a good one: that was also a highly extensible virtual world that really mostly got used as a social toy. It didn't *do* anything, and so there wasn't much reason to stay with it. Second Life *might* succumb to the same syndrome, but I'm not yet convinced.
The thing is, these articles are conflating two only semi-related aspects of a virtual world: extensibility and content. It talks about the fact that everyone plays World of Warcraft, far more than do Second Life, and seems to imply that the extensibility of the latter is the cause of that. But that's exactly backwards: rather, it's the *content* of the former that makes it compelling.
The problem is, extensibility and content are often (foolishly) made out to be in tension with each other. We have closed virtual worlds like WoW that have great content, and open ones like SL that don't have anything nearly so compelling. But it's ridiculous to say that this is an inherent problem of extensibility: that's like saying that Prodigy was obviously better than the Web, because they focused on providing content rather than extensibility.
He brings up the topic of VRML, and that sets off a lot of bells for me. I remember the VRML project intimately, having been deeply involved in it in the early days. (Hell, I came up with the name behind the initials.) VRML is a fine example of Missing The Point. Everyone got wrapped up in the bells and whistles of how to create the prettiest possible models and scenes, and ignored the few of us trying to talk about *context*. Discussions of how these worlds interacted, and what you could do with them, simply got lost in the shuffle. And so VRML became pretty much an irrelevance.
That said, just because so many people have done it badly doesn't mean someone isn't eventually going to get it right. It's not easy: you need to come up with an extensible system that allows creation of new spaces and objects; AIs; an open but manageable economic system; and the creation of *story* within that context. That's a big project. But nothing in it is impossible, or even infeasible -- indeed, much of it is solved problems by now. It *will* happen; the only question is when.
(This subject is near and dear to my heart -- it's basically the Braid project, one of the mammoth undertakings that are on my to-do list. I suspect someone else will get to it before I do, but we'll see...)