While I was an undergraduate, I had a class with a professor specializing in supercomputing... one with a grant from the NCSA, to study and refine (if I recall correctly) Lambda Calculus. Which was used, in her case, as a system for simplifying mathematical calculations which could, in the case of her prototype, be worked into some high-level optimization heuristics for High Performance Fortran.

By way of explanation, certain kinds of resource contention are very important in parallel computing problems - specifically network and network-like resources, as nodes in the computer may actually be slower collectively than an individual node if the problem they are attempting to solve requires too much communication with each other. Hence, optimizations that can automatically simplify out node interdependencies are very attractive to the community of scientists which make regular use of supercomputers; they're generally not the best programmers in the world to start with.