I also had this other idea once that didn't end up working but I think you guys might find it neat.
I setup a domain on an ip I don't use for hosting any other pages. Then I setup the server to read a cgi script as the static main page. The page was filled with content related to a random topic. Then at the bottom the script would read the referal(if no referal then it would pull from a master list). Then it would search overture for keywords and phrases like it. Then it would rewrite the http.conf and dns config for 10 subdomains with those phrases in them, and content generated from those keywords. Then restart the dns and post the 10 links at the bottom. IE.
http://greatwords.domain.com. Whenever the spider would crawl those pages the script would do the same thing again but with ten phrases found from that keyphrase. In other words it would go on forever, eventually spanning just about every topic on the internet.
This obviously didn't work very well, my site got crawled a lot but never ranked for anything, infact during an update Google hit that site 70,000 times in one day trying to grab it all, but only ended up indexing a couple of the sub domains.
Either way it was a fun experiment that tought me that the people who design search engines are obviously smarter than I.