The 90 deg pulsewidth is dependent on transmitted power and probe tuning/matching. A conductive sample will greatly spoil the probe tuning and matching, resulting in higher P90. One should retune and match the probe for conductive samples in order to bring the P90 back near its usual value. If the sample is so conductive that the probe tuning components have insufficient range to correctly tune and match impedance to the transmitter, then the higher P90 cannot be avoided. In this case, there is some danger to the amplifier due to excess reflected power.

Re-tuning the probe (correcting both tuning and matching) is essential, but does not necessarily restore the rf field to the amplitude obtained with a non-conductive sample. In my experience, conductivity affects the 1H channel most significantly, and the effect on 13C, 2H, and 15N is negligible. The effect scales with frequency, so I would expect changes also for 19F, and perhaps 31P to a lesser extent.

The relative loss of rf amplitude depends on details of the probe. Probes which apply a relatively high rf electric field to the sample (this depends on the coil design as well as the circuit design) will have higher sample-dependent losses. Probes with very high Q (e.g., cryo-probes) will probably be more sensitive to the electrical properties of the sample.

Here is a specific example: A cryoprobe with which I am familiar has a 8 microsecond 1H 90 degree pulse with non-conductive samples (e.g., chloroform in acetone-d6 standard sample). With salty protein samples, the 1H matching needs a large adjustment, and the 1H tuning just a little. The probe tuning and matching controls have sufficient range to fully correct the tuning (so the reflected power is negligible). Following that, the 1H 90 degree pulse is in the range of 10 to 13 microseconds, depending on the sample composition. 90 degree pulses for the other nuclei appear to be unaffected by the sample.

Conductive samples limit rf penetration. The probe will detune, but b1 will also become increasingly spatially inhomogeneous over the sample volume. The sample itself acts as a partial Faraday cage. The effect is more pronounced at higher frequencies.