Abstract+conclusions is the sweet spot.

This makes me think that given some particular article, the only people that would ever read it in its entirety are its authors and (maybe) a couple reviewers.