I'll have to think more about this question.
So far I've just been obtaining and comparing rise/fall delays (not rise/fall times, but rather the delay across the cell). That's the obvious way to measure it.
Watching whether the inverter settles at half-rail when gate is shorted to source would be a good solution, but from the spice tests I've been running just now it doesn't seem to exactly correspond to a balance between rise/fall delays.
Trying to experiment with these spice simulations is made difficult by the current state of the spice models (only the discrete ones being good)...