The following usage numbers are calculated by first averaging each host's data over a time period of one minute. Those values are then averaged over all hosts (still per one minute interval), i.e. we determine the average utilization of all hosts in each one minute period. In the final step we look for the peak number, i.e. the minute with the highest utilization. This is done for each value individually (CPU, RAM, etc.).
This algorithm limits the impact of individual short-lived peaks on the end result. It works best with many similar machines that are running similar numbers of hours per day.