Interpretation 1 is also mostly false. Unless you have a program that's already single-thread-optimized, that solves an embarrassingly parallel problem, and you have lots of cores at your disposal, ...