Note: All the examples below are highly old. The overriding problems were badly configured clients and/or numerous and almost synchronous ntpdate requests. Nowadays the situation is slightly different. NTP clients seem to be more well-behaved. On the other side, the number of pool clients steadily grows.
NTP traffic spikes and extreme request's values.
According to the documentation available on the pool.ntp.org, currently most pool servers get about 5-10 NTP packets per second with spikes a couple of times a day of 25-40 packets per second. Indeed, such a numbers are usually seen on the MRTG graphs with standard 5 minute average. The spikes occur whenever a server's time will come and the server gets into the pool RR DNS. Although the 5 minute average appears to be good choice, the question remains about the traffic behaviour on a more detailed timescale. Evidently, the values of requests per second smoothed over minutes or even seconds can sometimes be pretty much higher.
Two graphs below are not a pictures of any one-day traffic. Instead, to see more extrema at once, two weeks are used here in such a way that only the maximal '1 minute average' rps values from that period are plotted here (first graph). Similarly, the second and '1 second average' were used to plot the second graph.
A few examples of spikes and their fine structure can be seen here. The extremal values observed so far are presented on a separate page.
K. Sandler, October '06