Again, you misunderstand. If electric power input goes down, the overall heat from the system must go down by that amount, at first. The cold fusion reaction does not instantaneously jump to a higher level. It does not do this any more than a burning pile of wet green wood will instantly increase in power. These reactions have built-in latency, or "memory" as Pons described it. They are not stable but they tend to return to the same power level. The reaction might increase gradually as some control parameter is exercised.
Imaginary example for an e-Cat scaled device:
In the morning you have 20 kW input electricity + 20 kW anomalous heat = 40 kW total heat
Input power is reduced to 10 kW. Anomalous heat remains the same. 30 kW total heat.
Perhaps, for some reason, anomalous heat increases. After a few hours it is 30 kW. Total output is then 40 kW again. However, the total energy for the day will be lower. There will be a gap in total energy.
That is, unless Rossi has solved the control problem to an extent far better than anyone and he can instantly turn a reactor output higher or lower. And his control parameter has nothing at all to do with input power. (In other cold fusion reactions, input power is an indirect control parameter. Turn it off, and eventually the reaction peters out, after some latency delay.)
In some of his earlier tests, the reaction seemed to be real and it seemed to be on the verge of going out of control. That looked to me like a real cold fusion reaction. In the 1-year test, the machine produces incredibly stable output and an absolutely steady flow, which looks impossible to me. It looks fake to me. I doubt the instruments showed anything like this.
The flow is another problem. As Murray pointed out, given the geometry of the system, with some reactors shut down, there is no way the flow could be the same. The flow of water is cut off to reactors when they are being taken apart or replaced. There would not be EXACTLY the same flow every day to the nearest ton of water in these circumstances.