Some related projects:
Opis - http://citeseerx.ist.psu.edu/viewdoc/summary?doi=10.1.1.167.... - pure functions connected by dataflow combinators (with time travel debugging and model checking at the dataflow level)
Bloom - http://www.bloom-lang.net/ - temporal logic for coordination, message passing for communication with stateful endpoints
Flask - http://dash.harvard.edu/handle/1/2797447 - a c-like language for local sensor computations with a haskell dsl for specifying communication patterns across the sensor network
Then of course there is all the FRP work (eg http://www.haskell.org/haskellwiki/Functional_Reactive_Progr... http://docs.racket-lang.org/frtime/)
Despite all the interest, overlog (the predecessor to bloom) is the only language I've seen so far that has any convincing demos (hdfs and hadoop in 8kloc). I'd love to know if anyone is actually using this sort of stuff in production or, if not, why not? What goes wrong when you try to scale this up?
EDIT: I forgot to mention Mozart/Oz, one of the most interesting dataflow languages and one that explicitly talks about layering progressively less and less deterministic systems of concurrency - http://www.mozart-oz.org/features.html