A change of 2 pi radians in X Hz results from a delay error of 1/X sec. So a change of 360 deg in ~100 MHz is produced by a delay error of ~1/100e6 sec = 10 ns.
If data are vector-averaged across a phase slope, the amplitudes will be reduced, by a larger amount the greater the phase change in the averaging interval, so if the combined phase change on a baseline is worse in one polarization than the other, that polarization will have a lower averaged apparent amplitude.