> gain of a CF..., power tubes having a low mu, is actually substantially lower than 1 for a triode connection.
You can't derive gain from Mu alone (unless you make some assumption about Rl/Rp.
You can derive it from Rp/Mu, but that's the same thing as 1/Gm. So Gm is the simplest way to look at it.
For a given size cathode and grid precision, Mu and Rp vary all over, in the same direction, but Gm does not vary much.
So for a given load impedance, gain is a function of Gm which is a function of tube general size/class, but for "reasonable" tubes it will not vary much.
Also most of the promised benefits of CF operation go-away very quickly as gain goes down (as load becomes not-large compared to 1/Gm).
It can be convenient to define Rk as 1/Gm.
> use a pentode strapped as a CF, that would indeed bring you very close to 1
Pentode strapped as triode is just a triode. Possibly a better triode, because the main thrust of power-tube work was with pentodes; possibly not so good because a pentode can give low drop with medium G1-G2 Mu and a triode must have low Mu for low drop.
Pentode strapped as pentode does not work a lot better, Gm is actually lower, and driving the G2 to track the cathode can be VERY hard work.
> make sure you get the proper voltage swing....
In practice this problem is MUCH larger than it should be.
Take a triode CF with 300V plate supply. Best power is around Rl=2*Rp. Cathode swing will be about 200V peak. THD at large power will be around 5% divided by the feedback factor. Gain will be about Rl/(Rk+Rl). This will come out about 0.9 for many cases. Distortion of the CF alone is about 5%*0.1= 0.5%, nice. But grid swing will be about 1.1*200Vpk or 220V peak. A resistance-coupled voltage amp sees a light load looking into the CF, but still will make up to 5% THD when swinging peak voltages of 20% of the supply voltage. Or: the supply voltage should be 5 times the peak signal voltage for 5% THD.
So given 220V peak to the CF grid, we want 5*220V= 1,100V driver supply! We get ~0.5% THD in the CF, but ~5% THD in the driver. The "sonic signature" will depend very-much on the driver. We have just shifted our troubles from the output stage to the driver.
> You can apply exactly as much NFB, *or more* if you want, with amplifiers of much smaller signal capacity = lower distortion, with loop NFB.
That's the way I see it. The CF configuration only works for ~100% NFB, and for power stages forces an earlier stage to ABSURD signal levels. With plate-loading, you can choose 50%, 100%, 200% NFB, and the driver does not work significantly harder than the no-feedback amplifier.
- which needs to swing the full plate loaded drive plus the plate loaded output voltage... fun, fun, fun 
