I'd love to read some good experimental work on it as well. Off the top of my head, I can guess that there's some benefit to retargeting ads up to a point -- if you can show an ad sequence over time, there's probably an argument to be made that the spaced repetition is more effective than just showing the same ad over and over again, for certain kinds of ads. Then again, that gets back to ad design in and of itself -- a clever campaign could go viral and/or engage in a way an A/B optimized campaign with a poor concept probably never will. With something that's part art/game theory/design and part data, I'm really curious as to how you could even design a reasonably lossless experiment to accomplish this.