First, I don't deny that MS Office has a lot of power-user capability. Heck, for anything that's not in there you can always fall back on programming macros to add more functionality in. My point was that those capabilities are far from obvious, and seldom discovered or explored even by fairly serious users (like me). Think about it this way: Unix applications like vi provide an incremental learning curve; you start out only being able to do a little, but every bit of learning enables you to do a little bit more. Windows applications have a learning gap, rather than a learning curve; you start out being able to do quite a bit, but then reach a point where being able to do anything more at all requires an enormous investment of effort.

I don't think that one of these approaches is necessarily better than the other. Rather I think that to make Linux on the desktop appealing, you have to play to its strengths: harder to use, but very powerful, and with consistent pay-off for learning invested.

Second point: Your CRM example is an excellent one. I deliberately avoided the whole subject of middleware type applications. That's a complex subject all its own. But I do think that those applications reinforce my main point: purely Open Source software is not ready for the enterprise desktop, and compatibility with the proprietary world is a big reason why not. I will say that the number of enterprise workers who need access to clients talking to middleware apps is limited, and so again a hybrid approach -- mostly Linux desktops with some Windows desktops -- still should be the most efficient and cost-effective way to go.

Finally, it's important to note that web-based applications are making the middleware issue less of a problem for Linux. With previous employers I have comfortably used both SalesForce.com and CorporateTime from a web browser, and thus not had to worry about what the underlying OS was.