I think there needs to be a better thin client than the current web browser, but I think that client will be an evolution of current web browsers rather than a new app.

Here's how I see it playing out:

JavaScript libraries will build increasingly more full-featured UI libraries. That's happening already with JQuery and YUI and Scriptaculous and Dojo. The increasing complexity will put increasing demands on the client-side processor, so browser makers will respond with faster JS implementations and specializing JIT-compilers (like Psyco or the JVM).

Meanwhile, SVG support will grow - it's already supported natively in Opera and partially in Mozilla/Firefox, with Safari and IE support under development. Once that's widely available, it'll open up a whole bunch of new app categories - games, grapics apps, maybe even photo manipulation if JavaScript gets high-speed array operations and access to pixel buffers. When enough people start using SVG, I could easily see browser makers implementing their SVG support to take advantage of 3D hardware acceleration.

How long do you expect all of this will take? Will we see full SVG support in one year, two years, ten years?

It's here now (via plugins, development versions, etc.), the real barrier is adoption. If you can dictate your user's browsers, you can use SVG. Problem is that for something to be useful on the web, it really needs to be present without needing plugin installation on the vast majority of browsers. I see that as taking a couple of years.

Any good examples of people doing useful things with SVG that I can take a look at?

A lot are actually using the <canvas> tag instead, which is a bit more widely-supported in present browsers, though I believe it's a bit less flexible.

There's the aformentioned Doom clone (http://canvex.lazyilluminati.com/83/play.xhtml). A bunch of JavaScript charting libraries have sprung up recently - one I'm familiar with is WebFX (http://webfx.eae.net/). I did a real-time charting widget for my former employer that graphed stock prices in real time on your browser; unfortunately I can't show you that. SVG's also apparently getting popular in bioinformatics for viewing protein-folding and other biochemical interactions.

Doesn't this come back to the problem of ontology? Given the different possibilities, what makes you so sure SVG will come out on top? Can it exist w/o a standards org like W3C to promote its adoption?

The Doom clone was why I specified 'useful' in my last comment ;)