raph: I AM impressed by the results of cleartype. I feel they are doing something pretty cool in the mid-term to improve readability on screen. However, it sems to improve only the horizontal spatial resolution and I do not understand why they do not enable it on regular CRT screens (there is no reason it could not be possible. Unless the electron beam is not focused enough to get good results... Or too many different cell arrangements on standard crt screens. Who knows.). I really like this idea of trading off color resolution for spatial resolution. It reminds me of Avi Naiman's idea of trading off b&w resolution for spatial resolution. (yeah, well, I am currently reading his thesis :)
Steve Gibson doesn't blame you for being confused. His explanation is better than the one I was just about to type before thinking to do a web search first :)
That said, on a CRT there's nothing stopping you from cranking the horizontal resolution of the driving signal up. Of course, it will still get blurred by the analog signal chain and the shadow mask for subdividing the pixels into colors, but there should still be a win. A signal with infinite horizontal resolution, no matter how blurred in the analog signal path, will have the same edge sharpness as in non-aa text rendering, while preserving the smoothness and accuracy of fully aa rendering.
Of course, when you actually do the experiment you'll find that virtually no X applications can deal with nonsquare pixels. At least there were some Cobol apps that had 4 digit date fields :)