Amazon.com, Lulu.com, and OBE's POD programs will keep RPGs available for the forseeable future. Meanwhile, places like ENWorld, RPGnet, PnP, and so on provide places for people to get info, get turned onto new products, and talk hobby. Paizo.com offers more functionality than a hundred cardboard cutout booths in terms of supporting existing customers. Sales may be low, but product line "subscriptions" would have been unheard of 15 years ago, so I think the base has in some ways become more stable.
GMS has been wrong before, and will be again, and I think a lot of it, when it comes down to it, is his biases. I am not speaking of him personally but only concerning comments he has made here and elsewhere. First, he believes the industry provides the support needed for the hobby. In his view, if RPGs go out of print, the hobby belly-ups. I respectful disagree. I can think of several RPGs that have been out of print for twenty years or more and are still played, weekend after weekend. Second, he believes RPG players are looking for creative input from publishers, as opposed to what I would call resources. I think the opposite is true; I suspect most gamers rely on less than 30 pages worth of campaign fluff, but buy sourcebooks left and right because they save on prep time. Third, he believes RPG products are consumed. I believe that, by and large, they are read. I think sales are semi-independent of enthusiasm for playing. People buy RPGs books because they are cool, and they play because they have a group. They don't buy books that suck, and they don't play if they don't have a group that suits their needs.
Case in point. GMS's Tome of Secrets, a player's supplement for Pathfinder. I think he underestimated the sophistication of his audience. After it came out and response was not too favorable, he came here asking, "What do you want?" The answer, essentially, was, "A better book." ToS has few concepts or ideas you couldn't come up with on your own, plus some mechanics easily converted from 3.5, already available. The classes in the book were readily identifiable by "early adopters" as failing to meet Pathfinder specs for best practices; they had the lingo, but the accent was all wrong.
I suspect a lot of the "shrinking" in the RPG industry is simply that people already own many books. They would like to buy more, but they are more discerning than in the early Golden Age of the mid 70s, the boom of the mid 80s, the wild days of the early 90s, or the Gold Rush of the d20 scene. People aren't looking to buy, just to buy.
It's like graphics cards. They need to be replaced from time to time, which is your base sales. Then you can get more sales when new games come out that require them; this is your "network externalities." If everyone plays WoW, sales for cards that run Wow well go up, and ones that don't, go down. Lastly, you can sell a graphics card to someone by making a graphics card that impresses them in some way. Maybe it can run monster graphics, or maybe it just has fan control, uses less energy, and benchmarks about 20% better -- either way, there is a price point at which it will sell.
Plus, the d20 boom has a lot of publishers in the mode of overproducing; too many books, with too much art, sold at too high a price.
When paper becomes obsolete, we can drag out GMS's theory again and poke it with a stick and see if it growls. Until then, we keep on rolling.