Check if zero-cell partitions are a problem

[atgeirr]

In #448 we merged a temporary workaround that prints the loadbalancing by creating an array and summing it up over all processes just to correctly throw in parallel (in case a process has 0 cells).

Actually, the simulator should tolerate this situation and proceed with correct results. We need to check this and fix it if it is not the case. Afterwards we should use an gather for performance reasons.