Sites in search results that rank highly while serving visitors only ads are the target of a number of papers by Microsoft, submitted to the SIGIR 2006 conference held last week. Michael Hickins wrote about them in Microsoft: Noise Improves Search Results, and Jeremy Reimer adds more details in Microsoft Research Peers into Search.
Both offer nice overviews of the papers presented by Microsoft at the conference, which cover topics such as query analysis, user behavior, and personalization. Eight of the thirteen papers are available at the Microsoft Research pages. If you are interested in how Microsoft might be ranking, and reranking, pages based upon measurements of human behavior, you may want to dive into some of these papers. A good starting point might be Learning User Interaction Models for Predicting Web Search Result Preferences (pdf), which details their approach to seeing how users' clicks and other activities might help them predict preferences for web search results.