There was a paper published about a month ago [PDF] in which the authors had access to usage logs of PubMed and found that a high percentage of people clicked through links in the "Related Articles" section to find papers they were interested in. Mark Rabnett discussed this further yesterday.
As far as I'm concerned, all this shows is that PubMed's reliance on only ordering search results in descending chronological order is failing to locate the papers that people are looking for, so they have to stumble around Related Articles until they find something suitable.