The likelihood is the user searched Google before they tried Pubmed or ScienceDirect: "Ingelfinger Over-Ruled harnad" comes up with an OA version as the top match.
With OAI and OpenURL the OA version could be linked in as easily as the aggregators currently linked to by PubMed (although perhaps not as reliably, but then if you get a hit at least you know the version is accessible). While it would be nice for services to link to OA versions, it doesn't take more than 30 seconds to copy/paste some appropriate keywords into Google, which seems to do a good job of discovering an accessible version. Tim Brody Citebase Search: http://citebase.eprints.org/