Optimally universal parallel computers

Abstract
It is shown that any program written for the idealized shared-memory model of parallel computation can be simulated on a hypercube architecture with only constant factor inefficiency, provided that the original program has a certain amount of parallel slackness.