Agreed on all points and I have a potential answer to the last question:
One of the main reasons we chose to develop Weaver (http://weaver.systems) is that the existing graph databases do not handle dynamic graphs well. When the graph is changing underneath, the long-running graph computations either compute bogus results, or else performance suffers greatly. Weaver was designed from the ground up to enable ACID updates to the underlying graphs, even in the presence of long-running, consistent, time-stamped computations on top.
At a higher level, the entire NoSQL field is moving from loosy-goosy systems that cut corners and provide weak guarantees in order to achieve performance, towards databases that provide stronger semantics and high performance at the same time. I suspect we’ll see more specialized data stores pretty soon.