I swear that this existed, but I can't find it anywhere. It was a engineering blogpost where the author (a YouTube employee) talked about optimizing page load times at YouTube, where the unexpected result was that average page load times actually went up because more people with bad internet connections started using YouTube.

I would love to cite this in a talk, but I can't find the dang thing anywhere. It's the perfect case study for one of my points, but I've completely failed at finding it.