I think its because the web is a such a big and diverse place, that he assumes any idea that sounds good and is describable in one sentence is already implemented.
Well, I think such a book would be useful, and strikes me as a pretty obvious idea, with plenty of analogies in related areas (ie. zoomable user interfaces; code folding).
Currently, if you are reading math and hit something you don't understand, you can refer to other references (which may not use the same notation or approach things in quite the same way), and/or try to work out the details yourself, or get help from someone else. Working through yourself and looking at other sources certainly helps one learn, but can be very inefficient, and there can be a high energy barrier to success.
A 'zoomable' math text would require a lot of work to create, but the writing could be done collaboratively (or even partly automatically). There is some software to be written to implement this idea, but given the large number of computer and other scientists who depend on learning and using math, it is surprising to me that there hasn't been more experimentation with different ways to present it.
For what it's worth, we've already got wikipedia which does math pretty well. Even though it doesn't have the fancy-sounding zoomable user interface (which imho suck w.r.t. usability), it does offer plain old, get this, hypertext, which allows you to perform further exploration.
As for your idea, I don't want to minimalize it [because well, with ideas, you never know for sure which ones are good and which ones are bad (w.r.t. how successful they are) until after they are deployed], but it seems to me it would be a lot of work, with an end-result that might be totally crap usability-wise, and of limited interest to a large part of the population. Plus it would (for the moment) require a computer to read it. And computers suck as devices to read books on.
