By default long lived Netflow cache data is exported every 30 minutes. This will cause some high peaks in the graphs since the flow data is exported at one time. The following commands will help to reduce the peaks:
ip flow-cache timeout active 1 ( Breaks up long-lived flows into 1 minute fragments and smooths out the peaks and get a more accurate data representation on the graphs)
ip flow-cache timeout inactive 15 ( makes sure that flows that have finished are exported)
Hope this helps
That did it, thanks! Been wondering about this for months.