In case of academic research papers typeset with LaTeX, the source file is something you'd likely want to consider the semantic equivalent of HTML. TeX should be able to render the same document with different output constraints ("responsive layout"), but because of the architecture (TeX itself is fully Turing complete), it is pretty slow at re-rendering an entire document.
Part of the allure of a static document format like PDF is that you can, in theory, fetch just page 454 of 6000 page document and render that: with HTML, just like with TeX, you'd have to get and render the entire document to be certain that the layout won't change after you've processed the whole file.
The option to specify the target output size / dimensions at generation time is a reasonable option --- ISO A4 / US Letter, perhaps a target for smaller devices (though a 6"--7"+ tablet should be able to present most reasonably-formatted PDF documents reasonably legibly).
For anything smaller, PDF isn't really well-suited, and your better option is to go with a fluid-layout format such as ePub, .mobi, or, yes, HTML.
Having largely switched to bookreaders (eInk tablets), in large format with ~300 DPI grayscale screens, I strongly prefer fixed-layout formats such as PDF and DJVU to fluid-layout formats, for the spatial/cognitive reasons many others have mentioned in this thread.
My (and others') point is and remains that a spatially-fixed layout does serve a useful purpose for some documents. Including the 140 million or so published books and an even larger count of formatted published articles.
Yes, for short texts, dynamic flow within an HTML webpage is useful. Yes, for very small devices, virtually any format sucks and blows (this is a device problem, not an inherent PDF problem).
I'm not a fan of websites that dump what should be Web-formatted content as PDFs. But I'm also not a fan of the notion that everything should be an HTML document either.
(I've used various online document formats for going on 40 years, from raw ASCII (or EBCDIC) through roff/nroff/troff/groff, HTML, LaTeX, various flavours of Markdown, etc. I've hand typed out several books simply to have a suitable online digital format of them (I hope this serves to indicate my level of obsessiveness, if not sanity, on this topic). I'm a huge fan of Pandoc and its ability to take a standard markup format and produce a wide range of output endpoints (usually: PDF, ePub, HTML, plain ASCII text, though a few others may be included).
I'm also a recent convert to large-format eBook readers. And from that experience I can make two specific observations:
1. The behaviour of HTML and web browsers on an eInk device really sucks. Pagination and not triggering scroll actions with the merest suggestion of a hint of breathing on the surface is hugely underappreciated.
2. PDFs (or equivalent paginated documents, e.g., DJVU) offer an excellent reading experience on such devices.
I'm not a huge fan of the PDF file format, mind, it's far too variable and has too many surprises and vulnerabilities. As a reading medium, however, it's quite good, especially when produced with competent tools.