Better measurements

I tried out my own suggestion in the previous blogpost, to place the time measurements in the model itself, starting time being just before calling bench() on the first entity and stoptime being exactly when the last entity stops working. Compared to the old "experiment"-level way of timing things, I get about a 2.5 second decrease in times, which is obviously a good thing since this 2.5 seconds is completely irrelevant to my experiments.