Scholliers et al. have designed and implemented a mechanism that allows message control within an actor to be parallelized. For example, in a dining philosophers example, the philosopher requests two forks; the order in which the forks become available is immaterial.
The presented data suggests that benefits are visible for many process threads. What is not clear is whether one can also specify messages to the scheduler (for example, either two forks or a knife and a fork). This would make the proposed mechanism even more useful.
The paper should be of interest to programming language designers, as well as implementers.