Interestingly, when I look at the graph of the request rate against time, overlayed with the Ave response time, there are clear peaks in the average response time which coincide with troughs in the request rate.
Also it's clear looking at the graph that as the response time begins to become erratic, so does the request rate and it begins to not follow the straight line you would expect it to, consequently, never gets to the desired rate at the end of the test.
Why would the response time becoming erratic have any effect on the generated request rate?
(Presuming it is that which is causing it?)