What a such long pain we suffered.
After more than three years from Kiwamu's discovery, the PDH box 50kHz oscillation issue was finally solved.
This "weird peak at 50kHz" was caused by the oscillation of the voltage regulator (ON's MC7912).
As it imposed common noise almost everywhere, it screwed up transfer function measurements
like EricQ saw recently.
The negative voltage regulator (79XX) tends to get unstable than the positive counter parts (78XX).
The oscillation was removed by adding 22uF electrolytic capacitor between the output pin (pin3) and the ground pin (pin1) of MC7912.
This is indeed more than 20 times of the specification you can find in the data sheet.