1) The MDC decoder in a radio's receiver requires a certain audio level in the audio available to it, which is tapped before the high-pass audio filter. That level is equivalent to a deviation of nominal 750 Hz in a wide band channel (about 15% of max deviation), though from experience most radios' decoders will work down to an audio level equivalent to about 450-500 Hz (or about 9-10% of full deviation).
2) If you program a subscriber for narrow band and then transmit to a base station that is still programmed for wide band, you are effectively reducing the audio level fed to the station's decoder by half.
3) Theoretically, reprogramming a radio for narrowband causes a shift in the width of the IF filter's bandpass, and if that is properly done, it should compensate 1-for-1 on the reduced RF level of subscriber radios' transmitted PL. That is to say, 300 Hz of PL into a narrow band receiver should still yield audio at about 12% of max deviation, and the decoder should work.
4) On the other hand, if the station's receiver doesn't have an adjustable IF filter width, it may try to deal with narrowbanding by simply changing the amplification factor of the audio amplifier, and if that change is made after the high pass filter (so that it doesn't also goose the audio going to the PL detector), you're going to have a problem.
5) So the bottom line is that for any given combination of station hardware and subscriber hardware, you're going to have to experiment to see how narrowbanding affects PL decoding.
6) For what it is worth, virtually all of my experience involves Motorola equipment: Quantar and MTR2000 stations and Waris/Jedi/XTS or XTL hardware, and we've yet to experience a problem.