Regarding your "mostly seem not to be cousins" comment, I think people have an overly narrow definition of cousins where 23andme has a much more strict genetic definition. If you look at their FAQ, they claim that it's fairly accurate that if someone is marked as a relative, that's a relative. The degree of relation may be off by a generation or two, but it's there.
I find it interesting how many people, even those who sign up for the service and enable visibility on relative finder, heck, even those who fill out their profiles with surnames and extra information, either don't respond to contact requests or entirely dismiss relations dubbed by the system "likely 3rd cousin" as "impossible".
At the 6th, 5th, and even 4th cousin levels, especially going into the past with typically large families, we are talking about fairly large networks of people. I see a lot of folks say things like "I can trace my roots back 10 generations", when they really just mean one single paternal line, for example. How can you be sure that someone didn't jump in the hay with someone else in secret that one summer in 1807, or that some forgotten sister of a great-great-grandparent went off to America for 5 years before returning to her homeland, etc? With such large numbers of people, the likeliness of having huge numbers of cousins goes way up.
For example, look how many hops it takes to get to your 4th cousin:
http://www.wolframalpha.com/input/?i=4th+cousin
Can we all vouch for all the surnames, brothers and sisters, grandparents, great-grandparents, etc in a relation even as "close" as this?