I suspect it might work quite well in collaborative mathematics. Publish (internally, to your collaborators) when you get a new idea, have a rough sketch of a proof, find a plausible conjecture, think of a good way of presenting something, etc. Eventually, of course, you'd need to clean it all up and get the bugs out, but frequent informal known-buggy "releases" might be an excellent way to get from zero to a high-quality research paper.