With only a hint as to what this Pregel thing is about, my guess would be that Neo4J and CODASYL are focused on persistence/storage whereas Pregel is meant for high-performance OLAP stuff. Storage issues like replication are probably less of a concern.

Notice that they're submitting the paper to a distributed computing conference and not a database conference.

Ultimately I doubt anyone but the Googles of the world have a need for this kind of technology.

There was a time when nobody but the IBMs of the world had use for computers.

I work for a company with a 30 million node social graph. That type of data is becoming more and more common. Including things like Twitter where the social graph is effectively open to anyone.