I also want network transparency. Right now, I can run a Windows program remotely, via X and WINE. I can't do that on Windows proper without a whole lot of setup on a Windows Server (I've done it). Yeah, X may be slow. It may have crappy security model. But in the end, it's better than what MS has and better than what Mac has.
I could imagine something different. Go back to core principles, and write a server using current design paradigm and knowledge of hardware. Get the server side running, and well. Write an emulation layer that will allow existing X apps to use it. And then start porting from X to.. say, Ywindows. :)
(Then again, I wished that DRM existed for Linux, so the graphical subsystem could have its own top, memory manager, process scheduler, and such. Then that would have set up Linux to be the dominant force in computation. Alas, it never happened that way.)
Most GUIs are relatively static. They are not Crysis. Even if you hate X's drawing primitives and believe absolutely in the power of client-side rendering, there are huge wins to be had simply by pre-rendering stuff and storing it server-side as pixmaps, then using XCopyArea calls to show them on the screen. Sheesh, I was doing that since I was 18 years old, and I figured it out using just the man pages. It's not hard. It's not even particularly burdensome. It's just a case of "herp, derp, we don't want to think about our rendering code" from the toolkit crowd. Do we have to burn the entire display stack to conform to toolkits written the stupid way around?