Workflows are personal. The two factors that most strongly shape my own workflow is that my editor of choice is emacs and that I do not work in large collaborations. (It doesn't hurt that I like to tinker.) The emacs learning curve can be quite steep, but because I do so many different things in it, small increases in proficieny have a magnified impact. Also, I tend to be easily distracted, so the fewer times I switch between apps, the better. Instead of desribing things in gory detail, I just tried to give a general overview. If you have specific questions, or want to see something in action, swing by my cube, D-343. I'm usually happy to talk about such things.

The first thing I do is write an outline using [org-mode]. This is a a plain text draft of the paper, though I do write the equations using TeX. Once things are reasonably converged, I export my org-mode file to TeX, using built-in exporter. This converts my outline headings into LaTeX sections, among other nifty things.

[org-mode]: http://orgmode.org/

I start a git repository with the rough draft TeX file from the previous step. When I send a draft to a co-author, I make a branch for him/her, incoportate the comments when I get them back, and then merge this back into my master branch.

I use [AUCTeX] which the default emacs TeX editing mode. It has all sorts of goodies: easy math-mode editing, cross reference support, commands for creating environments, etc.

[AUCTeX]: http://www.gnu.org/software/auctex/

I do not have a solution I'm completely happy with here. I tend to make an ADS private library for each paper and then export to BibTeX occasionally. I've played with both BibDesk and Papers and not been all that happy with either (though I do like Papers for reading/emailing).