ZUI
Jeff Raskin's Humane Interface, the fabled ZUI, is nearly here. Whether zooming forward and back above a desktop will be tedium or if it's more efficient than a conventional GUI is yet to be seen - I think it would feel more coherent than a WIMP GUI, though. As I understand it, zoomable desktops aren't a significant change. You have the same widgets (scrollbars, drop-down menus, tickboxes) but there aren't windows.
A WIMP (left) limits the viewable document to the size of the window it's contained within. A ZUI (right) has no need to for windows as you scroll and zoom about a much larger surface. While documents have no need for windows they obviously still require an interface - but the main benefit to not having windows is that will never overlap or require resizing. As the picture shows the entire document is on the desktop so there is no scrolling within a window, instead you scroll your view (as it looks down upon the desktop). New documents appear beside the old ones rather than underneath them as in WIMPs.
Modern file managers such as Nautilus use the contents of a file as the icon. A text file will have it's first few words used as the icon; an image file has itself as the icon; a video might have the first frame as its icon. I haven't a clue what the icon looks like for audio. [examples of the icon being the file rather than the icon being the file-type].
"You just zoom in, and as soon as you can read the text or see the graphic details, you can work on them. Then there's no need for windows, which you are forever opening, closing, moving or fooling with."
From a distant view of several icons you zoom in; the icon takes up your whole screen and you can see the contents of the file; select a line and begin typing; zoom back out (the file is saved as you zoom out - earlier versions can be reverted to). I find this more coherent than loading a word-processor and opening a file, editing, saving, then closing the application.
The icon for a directory will show the files and other directories it contains within itself but at a miniscule size. Going to an icon 3 directories deep involves zooming in on several folders nested within each other.
In a ZUI list items have their relationship shown by the size (as opposed to indentation like most GUIs - I guess). You'd zoom in to see the barely legible children items when the parent would be much larger.
A ZUI isn't really a 3D interface (two and a half?), but then I don't think that a 3D interface would be best anyway. Letters will always be 2 dimensional and they are integral to almost every interface. Fixing the camera to always look down upon the desktop seems best. The few 3D interfaces I've tried just rotate text so it's always facing you.
I assume a ZUI uses mouse movement to scroll through X/Y while some other input device does Z. I'm not quite sure what device would best suit. Unfortunately most have learnt that the 'push away' mouse movement means moving up the screen (that Z equals Y). While using the mouse in a WIMP GUI most people's other hands are free. They don't tend to click the mouse and type at the same time so for most people the other hand is free - perhaps another mouse dedicated to the depth axis would work. Although that would be like patting your head and rubbing your stomach.