Rich widgets obscure the semantics of hypertext. Only the code behind the form knows what the widget really does. Contrast to HTML 5, and in particular Web Forms 2. These guys are extending HTML to capture what we really do on the web. As a consequence, client programs can (or, have a chance to) understand the meanings of hypertext documents from the web.
Example: you want to write a script to automate some remote bookmark service, as part of your mashup. But this service, unlike del.icio.us, has no documented "API". So you have to download its form, complete it programmatically, and POST an entity.
Case 1: The form uses Plain Old HTML. You're golden. All the semantics are right there for you to parse, or read. You identify the name of the text box where you stick the URL, and the name of the text box where you add a description. You compose the URL encoded form data, and POST it to the action URI.
Case 3: The "form" uses XAML + Silverlight plugin. An exacerbated case of (2).
Mike Dierken justly analogizes: "RIA is to user interfaces as RPC is to messaging interfaces". And notice, it's Rich Internet, not Rich Web Applications. These technologies pay lip service to the web, but they're not advancing the ball toward building more and better links throughout the web information space.
I'm having an ongoing email exchange with my friend Peter. He's convinced MS and Adobe herald a new age of Rich Internet Applications. He pointed me to this guy who's backed up a tanker to the Kool-Aid trough.
Sure, MS and Adobe have to sell something as the Next Thing -- what else have they got? But we've had RIA ever since Java 1.1 applets. We have Flash. We have <embed> and <object>. Do you really think what's been holding RIA back is the technology?
Users have voted with their mice, and they've voted for the web experience -- exploring the web information space using hyperlinks -- as far more important than whizzy UI. Ask eBay. Ask MySpace.