Different user interface configurations such as the standard Unity, Gnome, or KDE desktops, should be relegated to some sort of theme file that describes what assets to load and where to put them. Plugins should be used to supply the various functions. That way if you want a lightweight desktop that loads fast like XFCE, you can have one. If you want a more full featured desktop, or one designed to make the best use of screen real-estate for touch devices, you can have that too. I think E17 actually covers most of this, and it is highly optimized, and doesn't rely on 3D for fancy effects but can still take advantage of it.
But the important part is there will be one environment to target, and eccentricities/nuances won't vary like they do between the desktops we have now. The same should go for the file manager/Open dialog/etc that is used, it should be standardized and support plugins/theme descriptions as well. If I start typing a folder name in the window, and then enter a folder and back out of it, will I still be highlighting the folder name I started typing or will I be brought back to the top of the list again? As the directory is read, will the window dynamically display as it is loading in, and jump around when I am typing said folder/file name, or will I stay focused on that area?
I just want this to be the same on every desktop I use, so that I don't have to second guess myself if I'm using a QT or GTK or whatever else app. There can still be different toolkits, but if they are all targeting the same environment, they will behave the same and it will only be the developers that see the difference. If I want to open files with a single click, everything should pay attention to that preference, etc.
Maybe the solution is to extend the reach of the free desktop initiative. But we should be able to mix and match any desktop component, and every toolkit should pay attention to the preferences we set and be able to behave the same if it is specified.