David
My Peaberry V2 is 9000 miles away operating as W8TOD by VNC so I cannot duplicate your tests.
You mention however that you are using HDSDR for your measurements. You can get some fantastic performance numbers if you dial the RBW back, average the noise floor, and read MDS as a 3 dB spike above the averaged noise. Frankly I think this is a more valid methodology for DSP based receivers than is the ARRL method.
But the results will not correlate with the ARRL method. A 3 dB spike above an averaged noise floor on HDSDR will be perfectly audible but it will not result in a 3 dB S+N/N ratio at the audio output. In fact, if memory serves, there was no significant difference between S+N and N on an RMS noise meter under those circumstances.
If you are using output RMS noise as your measurement then I can only guess that something about the Peaberry mixer/LO layout produces less noise than does the Ensemble II and, therefore, "uncovers" the op amp noise to allow for improvements.
Warren Allgyer
9V1TD (and W8TOD through the magic of VNC)
Warren Allgyer