> Text flow... are you talking about inter-word spacing and rivers? My take on that is: full text justification was always a bad idea, only promoted for aesthetic reasons and to show off fancy layout engines that reduced rivers and other bad artifacts. It seems some publishers are doing away with it in new books.
I would say that mathematicians are extremely picky about the quality of their displayed formulas. Most of what can be produced by mathjax is not of acceptable quality. I speak about fine-tuning of kerning, alignement, etc. Having "reflowable" text is a nightmare for typography-inclined people: the flow is an essential part of the content and you do not want to lose it!
Everyone prefers better typography, but what qualifies as readable and acceptable for math? Generations of mathematicians learned from handwritten (including horrible scribbles on chalkboards), typewritten, and facsimile/photocopied material (like many Dover reprints). Current mathematicians grab scans of old math books from online pirate libraries. Math material on the web is mostly mathjax, and it's just fine. Math rendering doesn't have to be up to Spivak's standards to be just fine for reading and learning. And there's an even greater gap between "just fine" and "acceptable".
Back to the main points:
Epub is a better format for everything except niche uses (textbooks, especially math, papers, etc) where there's an argument for pdf. It's rather weak, but I understand the argument and I don't throw a fit when such material is only available in pdf. Most regular books aren't even published in pdf format anymore, for good reason.
There's no reason to convert epub to pdf except to print it — and why the hell would you do that? buy a used print copy — or to send to someone who doesn't have an epub reader — in which case you're better off giving them the name of an epub app to search for and install, so that they can read epubs in the future, rather than converting every epub you want them to read into poor quality pdfs.
Meanwhile, mathjax will only improve, along with publishers' general css competence.
Your comment makes no sense. If you want to read a math-heavy do and you stumble on unreadable text, your experience doesn't suddenly improve if you say "well at least I can get JAWS to play audio". There is no tradeoff.
I would say that mathematicians are extremely picky about the quality of their displayed formulas. Most of what can be produced by mathjax is not of acceptable quality. I speak about fine-tuning of kerning, alignement, etc. Having "reflowable" text is a nightmare for typography-inclined people: the flow is an essential part of the content and you do not want to lose it!