I did that experiment, and I'm not sure what to conclude because while a 120ms exposure in EZCap did return "noise" the values in the fit file are actually higher than for a "good" 120 seconds exposure?
I did another experiment, however, that I believe does eliminate SGP as the source of any error. I did a 120 seconds exposure (same filter, max gain in ASCOM properties) using Nebulosity 4 demo with ASCOM camera selected, and the values in the resulting fit file are similar to the SGP result. Specifically similar mean and stddev values.
Values are from fits liberator. All values are for max gain and via the lum filter, same target.
EZCap 120ms:
Mean 13,462 stddev 342
EZCap 120 seconds:
Mean 11,417 stddev 1482 (a usable image)
SGP 120 seconds
Mean 3,345 stddev 166
Nebulosity 4 demo 120 seconds:
Mean 3,318 stddev 302