I'm not sure what type of answer to give here, except *don't be tasteless*, and *do what looks right*. The obvious thing to do, if an idea is spanning multiple paragraphs, is to make it a separate proposition and cite it from within the larger argument. I've found how-to-write-readable-computer-code type guides rather useful in this regard.

The harder problem appears to be when to group separate ideas into a larger paragraph, for which it seems difficult to consistently defend any universal prescriptivist stance. Here's an example (yes, I understand that it is a bit of a low-blow to bring up something you wrote a long time ago): it has often confounded me why the proof of Lemma 7 on page 18 of [your honeycomb paper][1] spans three paragraphs. On the other hand, I was *really* glad that Lemma 8 was broken into various pieces, otherwise it would have taken me much longer to parse.

Maybe the golden rule should apply: *would I want to read this in someone else's paper*? Personally, I find it much more psychologically satisfying to see "only one more thing remains" at the end of a paragraph rather than at the beginning of the next one.


  [1]: http://arxiv.org/abs/math/9807160