Because hacker news doesn't use a database, the site is file driven and runs entirely out of ram. Changes are made in memory and logged to disk and upon server restart, the files are lazily loaded on demand to prevent the need to load the entire db all at once. This might sound crazy but it's actually faster (once loaded) and easier to program than a relational database.

As hacker news was a fun project for Paul, I'd imagine he doesn't want to touch a relational database with a ten foot pole; I don't blame him. You can get a long way without ever using a real database, files and in memory hash tables work fine up to a pretty decent amount of traffic. It only becomes problematic when you need more than one server.