Basic circuit theory.
One way to look at it: distortion means the gain varies over a cycle. To reduce distortion, you eliminate reasons why gain changes. But in a limiter you WANT gain to change. And there isn't any simple way to reduce the gain-change over a waveform that doesn't also make average gain-change almost negligible.
NFB sets an upper limit on the gain of a variable-gain amplifier. If the raw amp gain varies from 1,000 to 2,000 (2:1 variation), you can use 100:1 feedback and the gain will only vary from 90 to 95 (1.05:1 variation). You can use assorted tubes and get the same gain every time.
However a limiter needs a gain that DOES drop, not a fixed gain. You could apply feedback, and it would linearize the no-GR distortion, but you are giving-up some of your total GR range. And the benefit would be lost as soon as gain does drop, which may be where your distortion is worst.
The most fancy GE tube limiter seems to have a couple extra triodes feeding-back to the input stage. As far as I can tell they are "open circuit" (no effect) at zero and low GR, only passing feedback at the highest GR.
> what about the feedback applied between the drain and gate of a JFET, in order to linearize the resistance curves?
Yeah, that does work. I think it is specific to variable-resistance FETs. I've never seen anything like that on tubes. I think it mostly just cancels another hidden "feedback effect" inside the bulk of the FET. We think about "Gate-Source voltage", but at these low levels the pinch-point inside the FET is not referenced to the drain but to some point between drain and source. The voltage-divider sets the gate voltage to be more-like what is happening inside the channel.
There are other topologies where feedback can be applied. A current-steering layout gives two outputs: one gets louder while the other gets softer. The sum of the two is constant. You can sum them and leak it back to the input as negative feedback. This won't fix all flaws though.