Hm. Looks like the PDF sample has some light (auto)hinting and a bit of linear alpha blending and gamma correction applied (compare diagonal stems, e.g. on the

Hm. Looks like the PDF sample has some light (auto)hinting and a bit of linear alpha blending and gamma correction applied (compare diagonal stems, e.g. on the "A"), while the DVI sample doesn't. I suppose this is less about DVI vs. PDF and more about Adobe doing text rendering properly, while Evince is probably using Cairo, which doesn't.