Were there gains AND losses along the curve I'd agree with you; but given the displacement of the curve for a gain throughout it's entirety, I believe it's more than just noise. 3-7HP is significant for a change of this type, especially given how good the OEM's have gotten at not 'leaving much on the table.' And as you pointed out, there's SIGNIFICANTLY more than 3-7 up top....looks like a win to me. 'Course, I don't know the price.
As someone who works with measurements and data, noise and tolerances do not always show up as deviations above and below the true data.
24 inches of water is about .87 psi, or a 6% pressure drop from atmosphere.
The PMS system is 8.1 inches water, or .3 psi. That is 2% drop.
In theory there should be less than 4% power increase *if* the pressure measurements are correct. The dyno shows about 5% at peak.
IMO, what looks very "strange", is the low RPM increase. If the inlet system is a restriction, the power increase should be very non-linear. It would have the most effect at the highest flow, rapidly decreasing to no change at lower RPM.
For example, at ~82% peak horsepower, the HP difference is ~308 to 291. That is a ~6% difference.
The pressure difference was measured at 18 H2O compared to 6 H2O, or .65 psi to .22 psi, or 14.05 to 14.28, or 1.6% air charge pressure change. So the dyno said the power increased ~6% for an air measurement change of 1.6%.
When I measure power levels, I play heck to get within 5% without using a caloric standard or normalizing to a caloric standard. Using a good caloric standard and calibrating, I can get well within 1%. I doubt a chassis dyno is 5%, if a system with no moving parts and no friction is only repeatable to 2-3% and accurate to 5% without going back to a caloric standard. I see this every time I do measurements when I measure small differences compared to instrument tolerances.
You decide what those numbers mean for you, but for me they indicate an accumulation of typical measurement "noise" or variability. Don't take this the wrong way, but "rubber rulers" might be marked in a scale of .01 inches when the variability is .1 inches absolute accuracy at big numbers is 1 inch.
I'm not making any judgement about the product. I'm simply pointing out what the data tells me.