That said, I think Wave is VERY WEB. Which is exactly why it stands a chance of succeeding where Ozzy's attempts have failed. I looked at the protocol and it doesn't seem all that crazy or complex. Robots can be written in any language as long as they can read and write JSON over HTTP. Gadgets are written in HTML and Javascript and are deployed over the web. It appears that near 100% of the gadgets and robots they demoed are UNIX-style "small, well-written programs that do one job well" and they plugin to a more complex UI.
Distributed, micro-kernel applications -- like the web -- do not preclude big ideas and large host apps. Browsers themselves are incredibly complex. If Wave proves to have utility and people use it, I don't see how it could fail.