IF the measured COP is real and not Calibration Constant Shift, then we can indeed inquire as to why they get more successes at a particular loading.
That's the $64 billion question isn't it. Is that COP>1 real or not? And yes, if the COP>1 is real, than one would certainly want to investigate why.
But you say it's a calorimetry error, and there was no actual excess heat. But what artifact of the calorimeter would know whether the loading was 0.94 or 0.92 ? Why weren't there as many false reports of success at 0.92 or 0.96? (The plot doesn't indicate how many FAILED runs there were at each loading for the SRI and ENEA data).
I've gone through the basic process before, but to repeat... All F&P type cells have been designed to date with all penetrations for electrode and sensor leads through the top of the cell. Those penetrations are heat loss pathways. They allow calories to slip out undetected, which is why one has to calibrate. When you put 1W in, you don't actually measure directly 1W out. (Highly efficient calorimeters get quite close, and some people say that they therefore don't need to calibrate. But that just means they are assuming a particular value of the calibration constant.) To compensate for the loss, the calibration equation typically 'bumps up' the measured value to what it was supposed to be. Then that equation is applied to temperature data from a non-calibration run, i.e. an actual experiment.
All F&P type cell calorimetry has assumed a single equation is all that is needed. A temperature or temperature difference or voltage (as in Seebeck cal.) is multiplied by a factor with perhaps an added offset value. More complicated equation forms have been used too, such as the McKubre transfer function approach. But because of the location of all residual heat loss pathways in one concentrated area, heat produced closer to that area will lose a slightly higher fraction out those paths. Heat produced further away has a better chance of being captured. So if you take some of the heat from near the loss points and move it to farther away, more of it will get registered as temperature rise, i.e. signal, and that extra fraction is then multiplied up and shows up as 'excess power'.
So, to get apparent excess power, heat has to move from one location in the cell to another (we're talking closed cells here), or appear where it wasn't before at all (as in an open cell where the effluent gases recombine instead of exit the cell). That means (closed cell) that recombination has to start someplace in the area away from the loss paths to get 'overdetected', and that means that process has to start up and start operating. To do that the special active state has to form, which requires whatever it is it requires to happen to do that. Apparently, based on the data, in Pd systems that is facilitated by having a higher loading, which as I described previously would theoretically cause a more active surface to form the higher the loading. Specifically how high loading translates to more active surfaces, especially given the impact of additives, is the open research area. But it should be reasonably easy to envision that, with a fixed experimental profile. that would occur at a reasonably fixed place. Of course a little bit of 'noise' would be associated with that, which is why you would expect a spread and not an absolutely singular number for the initiation point.
To fully understand the McKubre figure you need to know all the experimental operation details, which might be in his report, but I don't think so based on my vague recollections at this time. You would also need to compare the conditions used by MIT and Caltech as well. And of course the materials themselves can have an influence, it's chemistry after all.
Regarding false reports...there are many more of those...all the 'failed' replications. The CFers are right in that the early replication attempts were not adequate in many cases, but the key is to recognize that means F&P were premature in their announcement. Normally, one does not announce 'ground-breaking' results via press conference until they are ready to be published by a journal. F&P were forced into it though because of University concerns over IP rights. But the onlookers didn't know that and expected that what they presented met the normal reproducibility requirements. Unfortunately they didn't, and F&P still had to play coy because of IP concerns, and that just ticked off a whole bunch of people, and we ended up where we are today.
What we learned since then is that a there are a lot of F&P-type systems that can give apparent excess heat signals. What we haven't learned yet amazingly enough is whether they are real or not.