Your server needs to be able to cope with the peaks, not the averages.

Thinking of "events per long time period" as "an average of events per small time period" is an easy shortcut to overloading your servers.