One of our 1000 APs recently began to stop responding to SNMP when throughput went above ~60 Mbps. I started running an SNMP poll of the "System CPU Usage" and I'm getting numbers in excess of 143,000,000 while my other AP on the same tower is around "5" for the same poll. I don't have a frame of reference for what those numbers actually mean but they seem way far apart...
Those two numbers are 142,999,995 apart.
That's probably not very useful diagnostic info, but it's all the help I could offer. :)
Did you looked at Total Frame Time Used ? I have AP's like this :
No gaps in the SNMP obviously. However if one starts maxing out "Total Frame Time Used" then SNMP stops responding and gaps start showing up in the graphs. That means it's time to split a sector or upgrade a backhaul or wonder if you should have been more strict about customer minimum link requirements.