When looking at the locally available licenses, I see four entries: "Mathematica", "MathKernel", "Sub Mathematica" and "Sub Mathkernel". Now "Mathematica", "MathKernel" and "Sub Mathkernel" are clear, but what are "Sub Mathematica" licenses used for? Are there parallelization constructs for the front end?
The basic idea is that if you're running parallel computations that require a front end (for example exporting graphics in a bitmap format,) you don't want your main instance of Mathematica generating the graphic since it would then become a bottleneck. So your sub-MathKernel would use a sub-Mathematica for that purpose.