[nsp] CBWFQ (in GTS) causes voice delay?

[Paul Plantinga]

Is this correct?

(IOS 12.1(5)T on 3640)
I am implementing CBWFQ for a low priority, bursty, TCP source and a strict priority (LLQ), approx. CBR, voice aggregate. 

It seems that the CBWFQ first clears the TCP queue after the burst has passed then FIFO queueing (of the voice traffic) resumes after the congestion has passed. So the voice traffic following a data burst is delayed significantly while the CBWFQ clears the low priority TCP queue that has been created during congestion. Ideally it should remain in 'congestion' (CBWFQ) state to ensure voice traffic is still prioritised out.  

Sorry if this seems a little academic, but it does seem to affect the voice traffic.

Thanks, Paul