This Linux Host Software Analysis report shows the process data collected from the /proc table by ESALPS. "top" would show the identical information. Each of the 10 servers has the same identical configuraiton, with an SNMPD agent to provide the data to ESALPS, and five AFS daemons running AFS. Note that each of the AFS daemons is reported as taking 60% of one processor. With 5 Daemons running, each linux is reporting that it is using 300%. With only 3 physical shared engines, and 10 servers, this means that the data is on average one order of magnitude incorrect.
This data is just one sample of the problem, but shows that the data is extremely unreliable and can NOT be used for understanding the capacity requirements of any given application.