We actually did some work on this a year ago.  Idea was to crowdsource traffic data using a combination of manual (i.e. SMS text) and automatic (i.e. smartphone [specifically GPS + accelerometer]) means as a basis for building dynamic real-time maps of traffic flow from the people actually stuck in the traffic.  

Worked out really well -- we also have benchmarks between A*, D* and D* lite for this scenario and extensions to A* to reduce computation and space overheads.

As a side-note, earlier this year I created a proof-of-concept for predicting traffic flow using the data available from the Access Kenya traffic cameras available online.  It was surprisingly accurate and useful given the low frame rate and resolution of the image data, working excellently between directly connected nodes (junctions) and moderately well between indirectly connected nodes.

So yes, proof-of-concepts seem to indicate there is something to be gained.