Hi,
Compiling the Linux kernel 3.2.12, I've seen this new feature: KCOPY: "High-performance inter-process memory copies. Can often save a memory copy to shared memory in the application. Useful at least for MPI applications where the point-to-point nature of vmsplice and pipes can be a limiting factor in performance."
Could Gambit easily exploit this feature to optimize message passing in multicore machines with several Gambit instances, at least under linux?