So in summary:
1) There are lots of older devices in the market which will not display properly prepared eBooks (and there's not even a way to properly prepare in the Kindle format)
2) Publishers and distributors can prepare eBooks with non-scaling images for equations and uncaught spelling and other typographical errors introduced in the OCR conversion process
3) some money can be made selling junk and in the meantime not providing usable PDFs in the market.
Mediocrity reigns!
There's at least one author preparing both a full-sized and a tablet sized PDF set to avoid compromising on quality:
The Feynman Lectures on Physics.
Maybe a few more years and ePub 3 with MathML will be properly supported - currently Chrome and IE do not support MathML (
Can I Use MathML?) so that leaves out a large chunk of the market. At least it is possible to generate credible ePub 3 with MathML from LaTex:
Recommended workflow? LaTeX -> ePub with math.
Maybe if ePub 3 with MathML is supported then the hoards in India will be trained to re-write the equations using MathML - introducing more errors; or maybe someone will develop specialized math OCR to generate MathML.
I also observe that even if the OCR'd text layer is poor in the PDF, at least one still has the actual image of the text and that is not at all always preserved in the ePub / mobi versions. It is worth the minor issues with PDFs to have books preserved in a portable manner.
Bottom line: Offer the PDFs for those who prefer to use them. The marginal cost of the PDFs is minimal.