There is a big difference between the web (HTML + CSS) and LaTeX (LaTex + styles) worlds in terms of culture, and that defeats the idea of "separating content from presentation".
On the web, that content and presentation are separate, that's a technically a possibility, and culturally an actuality. That is, while you have the technical tools to apply such a separation, the toolkit does not force it on you; you can choose to write html with in-place defined colors and font sizes and other style elements. That the standards are higher than that and one is expected to write clean structural HTML and specify style separately with CSS is just a cultural phenomenon.
In the LaTeX world (scientific community), separation of content and presentation is also technically feasible, but the culture that maintains it as standard is missing. Most LaTeX documents are written by scientists who just want the damn thing look as they prefer, and apply all nasty tricks that the system can offer to get at there. The more computer savvy ones engage macro writing to save them a few keystrokes, but that's far from a way of document authoring that keeps content and presentation separate.
Over the years I've written nearly 200 papers with about 70 co-authors, probably 75% of them done in latex (most of the rest in some nroff/troff variant). The process of latex authoring usually involves iteration, with different authors contributing different sections, then a certain amount of rewriting of each other's content. Papers typically have a page limit - often 12 pages. But you ignore all that while coming up with the first few drafts. A couple of days out from the submission deadline, someone (often me) starts to panic that the paper is 16 pages and needs to be 12. That's when you start going through and copy-editing, trimming non-essential content and rephrasing for conciseness. Only during the last 48 hours do you start to worry about tuning the layout, because you know from experience that it will likely all change in the last day. In the last day you're doing fine tuning, panicking that the paper is still 13 paqes and needs to be 12. Now you do a layout tuning pass, and it's now that authors tend to get into all the nasty tricks like negative vspace, tuning caption distances, and so forth. Even then, your co-authors are probably still modifying content without paying too much about the effects on layout. Finally, two hours before the deadline, you've got it down to 12 pages. Now you have to stop changing text without regards to layout.
Anyway, my point is that in the weeks/months of writing a paper, the messy mixup of content and presentation really only becomes relevant in the last two hours before the submission deadline. Up until that point, everyone is mostly using the same subset of latex, not worrying too much about the presentation part, because the coarse-grain paper layout is usually handled by the style sheet you get from the conference.
On the web, that content and presentation are separate, that's a technically a possibility, and culturally an actuality. That is, while you have the technical tools to apply such a separation, the toolkit does not force it on you; you can choose to write html with in-place defined colors and font sizes and other style elements. That the standards are higher than that and one is expected to write clean structural HTML and specify style separately with CSS is just a cultural phenomenon.
In the LaTeX world (scientific community), separation of content and presentation is also technically feasible, but the culture that maintains it as standard is missing. Most LaTeX documents are written by scientists who just want the damn thing look as they prefer, and apply all nasty tricks that the system can offer to get at there. The more computer savvy ones engage macro writing to save them a few keystrokes, but that's far from a way of document authoring that keeps content and presentation separate.