The amplitude of the loop transfer functions plotted so far are actually the square of the real amplitude. The problem comes from the way I treated data saved by the spectrum analyzer. Each file is composed of 3 columns: frequency, real part (a) and imaginary part (b) of the TF.  Of course amplitude and phase are recovered by doing:

Amplitude = sqrt (a^2 +b^2)

Phase = angle (a+i*b)

Due to an oversight, I had replaced the sqare root with the absolute value in the amplitude computation. This explain the unexpected behaviour (1/f^2 instead of 1/f) of the openloop TF around the UGF. 

We will upload soon new TFs measurements (taken by Yuefan and Marc on monday night) properly plotted.