I think that more generally when a document becomes sufficiently complex it is better to use a page-oriented, typesetting approach. You'll want to fiddle with where certain page breaks and figures end up, and if you do that you might as well stick with one canonical format.
You can view it as the paper paradigm. Alternatively, you can also see it as sticking to a uniform, canonical format. Getting fluid layouts right is a very difficult problem, and manually getting the layout right for a fixed format is simply a much better solution for this purpose.
Which is more than a little quaint, in 2012.
Right, and I have even seen page numbers in citations of theses. It wasn't common in my field at least, but it happens. For a journal citation, generally page numbers tell you where in the volume that article is, not the content that is being referred to. I think there is flexibility in citations—one can refer to a figure, table, section or even paragraph number with at least equal clarity as a page number. There is convention here, I agree, and my point in my original post is that the convention is followed dogmatically rather than pragmatically in academia.
Getting fluid layouts right is a very difficult problem, and manually getting the layout right for a fixed format is simply a much better solution for this purpose.
I have given this topic a lot of thought. Partly from working on an HTML document for a thesis, partly because I've worked in the area of cognitive ergonomics. At the moment our tools aren't great for this, but that's really our fault as software designers. Fluid layouts should be better than fixed ones.
Concerns about fluid layouts are new, a byproduct of current technology. Printing technology didn't afford us these problems. You can remember when it was commonplace to see websites that contained text as bitmaps, to force readers to consume content as laid out by the designer. There were sites that rigidly spec'd font sizes and browsers that allowed that. Today, the best web designers accept the fact that content will be consumed on different screens (sizes and pixel densities) by users with different preferences. They design layouts that can gracefully handle a large font size stipulated by the reader, rather than design the way that a designer would for print.
So what is the underlying issue with fluid layouts?
Well, we have text that relates to graphics, and as Tufte points out in some great books on the topic, it's particularly useful (and historically very common) to have the relevant text presented in conjunction with graphics. That's an obvious point, but with fluid layouts this can unpleasantly 'break.'[0] This is a solvable issue (even using current web technologies), in a way that can evolve beyond what paper can do for us.
I'd like to see a markup that allows me to designate which graphics are relevant to a block of text, and have those figures shown together. For example, if the display allows it, once you start scrolling past a relevant figure it could slide to the margin of the text and stay pinned alongside as you read relevant text for reference. Perhaps the reader could select other figures to pin, or unpin figures as well.
Think of the number of times you've flipped between pages of a PDF to read text and look at the figure described. Sometimes I end up opening the same document twice and putting them on my screen side-by-side. That's the kind of hack us human factors types love putting in slide shows about how we should be designing for users.
With high-resolution displays getting to mass-market price points, we are only missing the right tools to take technical documents to the next level. As I said, this could be done in HTML, CSS and javascript today, but it's not author-friendly. Even if the tools did exist to make this easy, it would frowned upon in the academic world for being different. But some fields are more progressive than others, and eventually we'll move past the page as a paradigm. I've got to applaud the OP for taking steps in that direction.
My futurist speculation: the move away from pages in the scientific community will happen at the same time that it will move away from traditional journals as an idea distribution channel. I don't think that's in the immediate future, though.
[0] In fact, this often breaks with fixed layouts too, except it's only really bad during document creation. Think figure placements in LaTeX. Fortunately the document creator has to do this work once, and then all of the readers see the exact same thing. Better tools for fluid documents would make things friendlier for the author as well as the readers.
I put my thesis together in HTML using Pandoc, customized to look a bit like a Tufte publication with margin notes. It had animations, anchor links when cross-referencing paragraphs, and in my eyes, made sense for a document like a thesis.
My committee members, on the other hand, were in consensus that a thesis should be a PDF with page numbers, not some multimedia document with hyperlinks. When I made a PDF[0] and submitted it to the university, someone even checked it to make sure the page numbering switched from Roman to Arabic at the correct point as stipulated by the submission guidelines (it didn't, hence my knowledge of their thoroughness).
Considering that universities today often don't even want a hard copy of a thesis, this is a world that is tied very strongly to the paper paradigm.
[0] Fortunately Pandoc did this quite nicely, though I did have to make a few changes to deal with citations. Incidentally, there is no good, cross-platform way to put an animation in a PDF.