I measured the IMMT1T QPD2 spectra when PRFPMI was locked with 1W and 10W (fig1).
Red and blue lines represent the spectra when PRFPMI was locked with 10W and 1W, respectively.

Measured noise floors are about 1.4e-7 and 1.8e-8 with 1W and 10W, respectively.
So, my calculation for 1W (1.8e-7) seems close to the measured value while that for 10W (4.4e-8) seems about 2.5 times larger than the measurement.
I'm not so sure the reason but we ca obtain better SNR by increasing the laser power, anyway.