[CentOS] snmp values for interface traffic - Centos 3.5/x86_64

Is there a problem with the values returned from snmp for ethx interface traffic on x86_64? I have cacti graphing several servers that should peak around 200k/s and it sometimes pegs at 100M for periods of time when in fact the interfaces are not busy at all. I did a 'yum update' and rebooted, and things cleared up for a while, but now at least one is going crazy again, so perhaps the reboot just helped until something triggered the condition again. Stopping/starting snmpd doesn't help and the values from ifconfig look sane. I haven't seen anything like it on x86 servers doing otherwise similar things, just the x86_64's.