I made a mistake in the evaluation of P4 differential noise, because I was assuming the impedance in LN1 to be 228 ohm as for P6, but in fact it is 128 ohm. In fig 1 the new evaluation of the differential noise (in agreement with the lab measurement).
The conclusion about the limiting noise before unplugging the coils does not change: differential noise can explain the excess noise. This is shown in fig 2: adding in quadrature the differential noise to the residual (unplugged) noise, we obtain the'plugged' noise.
The increased series resistor should put the differential noise at a negligible level.