I am revising a paper now and one of the reviewers sent their comments as a PDF what looked like a Word document or RTF containing the LaTeX, rather than a rendered version. So it was a little annoying to read, what with all of the $’s and so on. The beauty of it was that I could just cut and paste from the PDF into my document of responses to the reviewers without having to reformat and it (almost) rendered without a hitch. There were some lingering issues though with quotation marks (I hate smart quotes) and itemizing/enumerating lists.
As a recommendation, I think that the raw .tex file of the review should be uploaded instead. That will make it much easier for the authors to revise, no? I plan on doing this in the future. What do you do?
An Ergodic Walk 
To some degree, I use (and welcome others to use) LaTeX typesetting in emails or other text-based communication (such as manuscript reviews), and can generally read the raw LaTeX markup just fine. Perhaps this is due to my upbringing of writing TeX in emacs, rather than using a WYSIWYG editor like LyX (though the idea of WYSIWYG TeX editors seems self-contradictory).
Obviously, past some level of complexity, it becomes exhausting to do the work of a LaTeX parser in your head, so I would hope the referees would have taken the time to compile the equations in this case, but maybe it’s not so bad to just cut-and-paste the equation into something like LaTeXiT and render it yourself. You can think of it like the modern-day equivalent of Invisiclues, or those answers that were hidden behind red hash marks and required red cellophane secret decoder glasses to see!
Oh I can read the raw TeX just fine, but sometimes the formulas I use get a little baroque and it’s nice to be able to render it out. What I was thinking is this : referees send you the raw .tex, you paste it into a doc which is your response, and then you can immediately render it and then write your responses as well. It would be much faster than cutting and pasting from a PDF and then futzing to get the math to look like it does in the PDF.
In total I think it would save maybe 1 hour of work, but it’s an hour of obnoxious work that gets in the way of actually, you know, revising the damn paper.