Joe Hewitt would like to see the death of the application in Longhorn.
We have been involved with the OpenDoc effort back in the mid-1990s. Moving to a component-oriented architecture is an incredibly complex problem to tackle. Both from a technical (plumbing, component-interaction, UI) and business (support, business-model) perspective. It would require fundamental changes in the way platform vendors and application developers do business.
While I agree with Joe that this would be a bold move on behalf of Microsoft, but I don't think they would be able to pull it off. Robert Scoble's approach seems more pragmatic: Making applications work together seamlessly. And it's a tiny (but important) step towards a more component-oriented architecture.