This article does a good job explaining some of the shortcomings of the UDDI architecture and makes some suggestions for improving the situation. I just don't think that the suggested improvements will fix the problems.

I see two significant problems. When I'm looking for a pair of pants, the inseam can vary by an inch, but I want the waist size to be exact. I need a way to say that, either in the advertisement or in the query. OWL is an excellent framework for someone to build the ability to infer that a "car" is an "automobile". However, neither UDDI as it exists today nor OWL provides a concept of what constitutes a match.

The second significant problem is the absence of a way to specify a business process. Only the most trivial web services consist of just a WSDL request and reply. The interesting ones consist of a "conversation", a series of interactions that often take different paths for different uses of the same service.

A minor point is tModels. They're a wonderful idea, and I wish we'd invented them for e-speak. The problem is that there's no standard for describing what the tModel stands for. Hence, you either know what one means because you've seen it before, or you assign a person to figure it out.

Qualifier: I am one of the co-authors of the Version 2 specification, and I haven't kept up with recent work on the standard.