The MQPerf probe aims at identifying a performance optimum. The tests most probably use up all of some system resources, so are very likely to be impacted by any external event which would need such resources. For example a test being performed on a workstation, while the programmer executes some heavy tasks, will result in oscillating production and consumption curves, with possible deep holes.
Using a Windows client workstation is a special case. You may notice random freeze periods on such a system, for a 5 seconds duration. Such freezes naturally impact the MQPerf tests. They show on the production and consumption curves as 5s holes, and also impact the system indicators in the MQPerf standard report, notably the number of pending messages.
The holes do have a logical explanation, but we can draw the following conclusions from this experiment: