I'm with BrowserUK on this one. Unless you can specify how big your problem is space/processing wise you cannot make a definite recommendation as to what type of solution to use. That would be as irresponsible as recommending some kind of flat-file tool for 1K users planning to perform read/writes in real time.

Those requirements exclude common relational databases...

I cannot say that this statement is definately true. There is multi-master replication available for PostgreSQL clusters up to 128 nodes. I'm fairly certain that other clustering relational databases have multi-master models nowadays as well.

Also, after reading a bit more about how this Raik tool stores its data it sounds an awful lot like you might consider Oracle's Berkeley DB which appears to already have a Perl interface written for it.