It is a very interesting topic what the ideal load might be for triodes in push pull. The better explanation I could find for this when I looked at the topic some years ago, was in a book from the staff at MIT. If you look at this article from about page 7 on, there is an explanation of it based on their reasoning.
One interesting thing is that the ideal plate to plate load resistance in push pull for the particular valve depends on its bias conditions, whether it be push pull class A (a rare configuration these days but not in the 1920's) push pull class AB or push pull class B because in those cases one valve has a significant to little effect on the operating load seen by the other valve over the range of those bias conditions. In any case it makes for an interesting read:
http://worldphaco.com/uploads/UX-171...amplifier..pdf
Also, an Ultra-linear connection pushes a pentode part way towards a triode in its operating characteristics as David pointed out.
Designing amplifiers with triodes there was always more of a concern to use the anode resistance specification of the valve as a parameter to help calculate the ideal load resistance (as pointed out by the staff at MIT) however, with pentodes they tend to act more as an ideal transconductance device, so other factors can dominate in the calculation of the ideal load resistance.
As always though, the same for triodes too, it is largely determined by the required maximum power output from the available supply, while still keeping the valves inside their maximum ratings, combined with how much distortion one is prepared to tolerate.