In general, non-linear devices are semi-linear to a point and then they distort bad.
The maximum undistorted voltage across an FET which is not cut-off is about 100mV.
If we build a compressor to do 20dB or 10:1 GR....
series: maximum input is 111mV, with 100mV across the FET and 11mV output.
shunt: maximum input is 1,100mV with 1,000mV across a clean resistor and 100mV output across the FET.
The advantage is more compelling at higher GR.
In fact with the FET on the compressed side of the action, there is no real maximum input. We could put 100V into 1Meg and a JFET for 60dB GR at 100mV output. We don't, because the idle noise of the 1Meg may be audible, and we don't have 100V sources, nor a real audio need for 60dB GR.
Inverted, we'd first have to pad 100V to 100mV, then for 60dB GR our nominal output has to be 60dB lower or 100 microvolt, an annoyingly small signal.
Photo-resistors can handle more than 100mV cleanly, though at 3V drop the disortion may be audible.
Junction diodes are clean up to maybe 10mV. This was good enough for one of the CBS boxes, because the node impedance could be very low.
Some old-old varistors had an "undistorted" band of several dozen volts. This was sloppy for protective devices, their main market, so they have been "improved" for much sharper knees. Your modem does not burn-up, but the old varistor tremolo scheme can't be built today.