Disclaimer
The Author of this posting offers the information contained within this posting without consideration and with the reader's understanding that there's no implied or expressed suitability or fitness for any purpose. Information provided is for informational purposes only and should not be construed as rendering professional advice of any kind. Usage of this posting's information is solely at reader's own risk.
Liability Disclaimer
In no event shall Author be liable for any damages whatsoever (including, without limitation, damages for loss of use, data or profit) arising out of the use or inability to use the posting's information even if Author has been advised of the possibility of such damage.
Posting
Could be transient microbursts. These won't normally show in "routing" monitoring.
Since you mentioned you're doing QoS, QoS divides up its buffer resources to service 4 queues per port instead of just 1. This can lead to insufficient buffer space for some queues. This might be mitigated with some manual buffer tuning.
You also mention using Etherchannel, but are you getting optimal link balancing?
As to limitations on the original 3750 series. You can run into performance issues if you exceed hardware resources, i.e. TCAM. This is best avoided using the correct SDM template. (NB: -12S offers a template unique to it.)
Another performance limitation is StackWise. StackWise puts all taffic on the ring and it's the source that takes it back off the ring, not the destination. (NB: Stackwise+ on the E/X series is "smarter".) Stacks with more and more members can overload the stack ring. Stackwise is spec'ed as 32 Gbps, but its actually dual 8 Gbps, duplex.
