It appears to me that the control system feedback loop was making improper assumptions about the throttle valve response (either because the assumption was plain wrong in practice or because the valves themselves had some sort of issue responding as expected to the control signal).