Because of Fitts' law, it seems that positioning commands at the immediate proximity of the cursor makes it possible to access them relatively fast: probably not as fast as elements in the corners of the screen, but still much faster than buttons which are positioned arbitrary.
Microsoft Office, for example, used that to display a standard menu with controls such as bold or italic buttons or text color selector very close to the cursor, the menu being shown when the application believes that the user may be interested in those commands. Given that the menu appeared when the user doesn't need it and remained hidden when the user needed it, and given that small controls were jammed together in a rectangular-shaped zone, this element was probably not the best example of successful user interaction, but still a partial illustration of Fitts' law.
At the same time, there is an interest of UX community in radial menus, as shown when searching in Google Images for “ux selection wheel” or “ux radial menu.” But the examples of the menus I've seen are more about visual design and much less about user experience: they are just a pretty way to group the elements together, but don't seem to improve the productivity of end users.
What about combining those two ideas, that is using radial menus and positioning those menus around the cursor itself? Actually, the two usages I know where button wheels were combined with the idea of immediate proximity of the cursor are both are very domain-specific:
Microsoft Productivity Future Vision videos:
It shows a very specific case of tactile interface, and while Fitts' law applies and there is a motion constraint (that is, moving the arm in such position is a difficult and slow task compared to moving just the fingers), it's still different from a cursor (especially because fingers, in this configuration, are much more precise than a cursor which may move too far away from its original position).
Splinter Cell video game:
This second example concerns a first person shooter, where a cursor has a very specific role of moving rather than pointing at an element. One aspect of that is that direction is much more important than distance: I haven't seen the actual implementation, but I'm pretty sure that if the person looks to the left, the yellow weapon will be selected, and will remain selected no matter how long the person continues to turn left.
Is the “if you move too far away from the original position, you're screwed” aspect of cursor-based button wheels the only reason which prevents this element of user interaction to be used much more than it is used today?
What could be the other caveats which prevent global adoption of this element in desktop applications where the primary interaction is done through a mouse?
It looks like this question is on a different subject than Are circular menu / button interfaces intuitive? The linked question is about radial menus as a form of organizing buttons in general (including on tactile devices). Mine is about radial menus positioned specifically around the cursor, in a specific situation of a product where most interaction with it is done with a mouse. I already addressed the aspects discussed in the linked answer in the third paragraph of my question.