I assume you have found John Broskie's essays on the
White Cathode Follower.
An objection is that John treats it only with small-signal approximations. The amount of bent-ness as you approach clipping is not clearly revealed.
If you just want to optimize load impedance:
Design a one-tube transformer-coupled amplifier. Pick an operating point. If you are abusing a WCF to avoid transformer step-down, you want to aim for lower plate voltage and all the current it can stand.
As in any triode power amp: best power is generally Rl=2*Rp where Rp is estimated at the proposed operating point (not the book spec). But distortion is very much less with Rl=10*Rp. Using Rl=5Rp is often a good compromise between power and distortion. If you can't get the load impedance that high, go as low as you want: distortion does not rise much as Rl goes toward or below Rp.
When you find a good answer, use twice the supply voltage, stack two tubes, use half the value of Rl that you found in the 1-tube design.
But as you go to different impedances, or toward clipping, the WCF gets bizarre. Optimizing different performace aspects gives different values of the top resistor. As near as I can tell, they are all wrong; the WCF can never be made "optimum". It is the most contrary "simple" circuit I know. (The SRPP may be worse, but I pretend it was never invented.)
Here is a not-bad WCF simulation, one I have looked at many times in many moods. CAD is often wrong, but when this sim doesn't crash I believe its curves are approximately correct. This happens to be two 2A3 with 400V supply idled at 55mA driven with 70V peak signal. Both plate currents and grid-cathode voltages are plotted against output voltage with three different load impedances. Note that the horizontal scale changes. I should re-run the series with fixed scaling but too tired.
Note in particular, the way it works push-pull in low-Z loads but push-push in hi-Z loads. When load is about Rp or higher, you really ought to look at a single cathode follower, one less tube to heat.