The Baconizer database contains 472,772 books, CDs, and videos spanned by 3,787,049 links. According to the FAQ these are taken from Amazon's links to related products. It then uses these links to find an association between two items. It's the sames as the 6 degrees of Kevin Bacon but with products instead of people.
These queries are made against a graph. What I'd really like to see is if an RDF triple store could beat the Oracle database at these kinds of queries.
Here's an example of "Practical RDF" to "2001 - A Space Odyssey".