[Reprinted from a Google Buzz post of Feb 12, 2010.]
As a professional courtesy, research papers in mathematics should be at a “local maximum” with respect to the results-to-effort ratio: any “cheap” consequences, generalisations, variants, illustrative counterexamples, etc. of one’s main results should be put into the paper if this can be done with only moderate effort on the author’s part. If one is too lazy to do this, these consequences might not appear in the literature for some time (as they are too close to your own paper to be separately publishable in their own right), and each reader may have to rederive them by himself or herself, which is a much less efficient process in the long run.
Conversely, if a huge fraction of the paper is devoted to only a minor extension of the main results, one may consider removing that section, or replacing it by a sketch or even just a remark; it may be that a subsequent paper is able to achieve that result with much less effort anyway.