Roger Cicala of Lensrentals.com has published an interesting blog post
with the results of his tests of sample variation in lenses and camera bodies (mostly Canon).
Using the Imatest software analysis suite, he tested the variation in sharpness among different copies of the same model of two different Canon macro lenses and a similar focal length Zeiss macro lens. He tested at least six copies of each lens, and in one case more than 20 copies.
He also tested the accuracy of autofocus compared to highly magnified Live View manual focus, as well as the repeatability of autofocus using a single camera/lens combination, and the variation when using multiple copies of the same lens model on a single camera body. Finally, he tested multiple copies of a particular camera model (the Canon EOS 5D Mark II) with one known lens.
His results are presented in useful graphs; they show that sample-to-sample variation is easily detectable in the products he tested, but its significance for real world photographs in typical end uses may not be particularly great.