Pay close attention to the unit of measurement being used in different views of Performance Advisor. For example, when looking at the "Overall Latency Per Protocol" view for the controller, the CIFS counter is shown in miliseconds (you'd probably see 3ms here). However, if you drill-down to a individual volume that may be used for a CIFS share, and look at the "Overall Latency by Op Type", they are reported in MICROseconds, not millieconds. So you'd see something like 1,000-2,000 microseconds (1-to-2 ms) for the actual volume.
This isn't necessarily a bug, but its been requested that future versions of Performance Advisor show everything in milliseconds for consistency.