And this applies to HTML5 video: Theora is cheaper than h264. So Gruber's argument doesn't seem to hold.
For example, MySQL wasn't and isn't as feature rich as the big closed databases, but it solved many problems as well as they did, and in many cases with less overhead (learning curve, administration, hardware requirements). So, in a way it was better than the alternatives, and that's why it gained traction.
Compare to H.264 vs. Theora. Theora is just an alternative, it doesn't solve any problems better, it is just "open", which most people don't care about.
I don't recall either gaining any traction until they were better in some way, either for developers or implementors. In some cases just being "open" makes it better, but that's only true if the project is going to be adopted in some way, like Apple using KHTML for WebKit.
In the case of Theora, nobody wants to adopt it as a long term project, they just want to implement video for their users. From that perspective, H.264 is better (I'm a layman here, but as I understand it the quality is better and the file sizes are significantly smaller, but correct me if I'm wrong).
It would certainly be better if they all got together on Theora much like they have on WebKit, but it seems ongoing development can't have the same level of improvement on a video codec that it can on a rendering engine, at least that's my perspective on it.
No, when they got traction was before they were better, but after they were good enough, which means you could run your business on cheap Intel/AMD hardware.
Theora is better in terms of cost, but that apparently isn't enough of an advantage to counter the technical shortcomings.
The lack of an alternative choice will push a lot of people into the early adopter space for some product niches, which can lead to a fast elevator ride into mainstream usage for some products (c.f. MySQL, Linux, etc.) This is not the case with video codecs. The field is technically rigorous so only a few have the chops to contribute code, the alternatives (h.264) are currently better and widely available so few have much to gain by building the Theora ecosystem, which really only leaves the evangelists who are interested in the product as a mission rather than as a tool. This group is unlikely to gain significant traction with the mainstream computing audience.
But, just so to make that clear, cost does of course not win by default. If something is bad enough, paying for the alternative could be the better thing to do.
And yeah, cheap doesn't always win if it really sucks. You could make an argument that Theora sucks that bad, especially when you consider the lack of hardware decoders. But that isn't Gruber's argument at all. His analysis here is really off base.