My gut tells me that none of the approaches I know for apparent concurrency are going to work well for a highly concurrent application on highly concurrent hardware. If that is the case we will see something new and different as soon as the cool hardware gets into circulation. (via James Robertson)On the other hand I don't see everyone cutting rope and heading over to Erlang. Probably some languages will evolve concurrency mechanisms. Those that are able to evolve most easily will be the simpler dynamic languages like Lisp, Smalltalk, Python, and Ruby.
On the way I would expect simpler extralingual concurrency mechanisms to gain popularity, e.g. JavaSpaces, esp. in the form of cross-language tuple spaces.