We have a the following structure:
Internet->CISCO->MTRouter->Switch(MT PPPoE AC, DNS, HTTP, MAIL)Switch->MT Bridged WDS Node and AP->MT Bridged WDS Node and AP->MT Bridged WDS Node and AP->MT Bridged WDS Node and AP->MT Bridged WDS Node and AP…
I still looking, but no lucky on searching where is the leak. 
At the MTRouter we have simple queues and queue trees. We have a 20mbits link to the internet. When the link go up to 14mbits, people who are playing games eventually got high ping times and get kicked. At this moment, we see that the pps/traffic drop down on all bridged wireless nodes. I thought that there should have peaks instead a hole when this happen. On the MTRouter everything seems ok, no cpu peaks or memory. The only thing I could identify is that when the clients ping time go about 2000ms(most of all clients connected, about 5 leaks per minute, p2p closed, 65% main link usage) every wireless node have a small hole in the packet flow, hard to open web pages. We used to shape p2p traffic from times to times to help. We have about 1000 wireless clients. In normal circunstances, every ping time is about 5-20ms (half clients connected, no leak, p2p open, 30% main link usage). Cant find where is the problem, please help.