SimEvents blocks' statistics are updated not continuously, but rather when an event is triggerred. The system state between the events remains intact. You can observe that the right scope is updated only at times when there is a change in the left scope, which indicates an event.
You should also note that statistics like average utilization are mostly used asymptotically, that is, their value after a long run that converges to a certain value is considered, not values in different instants. This is because that such statistics divide the amount of time the resource is used by the total time. Therefore, the simulation must run long enough in order for the statistic to converge to a number. You can observe in the right scope that the plot is converging to the 1/30 fraction you mentioned earlier. If you run the model for a long enough time (say, 5000), you would see that the last value of the average utilization is very close to 1/30.
As an example, here is average utilization for one of my models. There are lots of fluctuations at first, but as time passes, the plot is damped and converges to more stable values. You should not consider initial values because they are the result of few observations and cannot represent the long-term performance of your system.