It’s not that big:
Of the sixty articles in AER, EJ, JPE, and QJE that had been cited more than 500 times, only one article contained an author written lemma.
Here is the paper, and thanks to Johan Almenberg for the pointer. On p.6 you see a good graph, "Lemma frequency by decade." Note also that a top paper in Econometrica is much more likely to have a lemma.