what's going to be fastest driving a load? 1, 2 or 3 stages of inverters. I would have previously said 1, but in this case it's 3

If you want the quickest estimate, try to have the ratio of sizes be always ~3-4.

These slides may be helpful.

What is the load? Another minimum size inverter? The three stage inverter is better for driving a larger load. I think that the best way to characterize inverter delay is with ring oscillators.

Btw, I did FO4 delay here: https://github.com/20Mhz/spice_decks/tree/sky130/calib as hinted above and mentioned in the slides, best case is having geometric mean as fanout. cheers

"Logical effort" is a technique I remember learning about for performing back-of-the-envelope calculations on exactly this. There's a Wikipedia article on it here https://en.m.wikipedia.org/wiki/Logical_effort