One more thought: I draw my initial AIW in the classic late apex manner from an overhead view. This will get the AI to an 80-85% performance level because it does not take into consideration rough spots in the track, up/downhill terrain, passing & blocking strategies, nor handling characteristics. Subsequent iterations of tweaking waypoints and corridors then get you to the 95% mark where you start looking at specific mods. If you need the AI to get orderly as they come into a chicane, gradually tighten the corridor in the braking zone.
As I've been making Continental Divide Raceway (CDR) AIW, I got to the 95% mark and started in on the other mods. Historix is an obvious choice with a historic track, so the TC65 sedans were high on my list. They all fell into line except the Mustang, which was way slow and far too tail-happy. I was beginning to feel like it was just the way things were. Then I had an aha! moment while watching the AI and, after a few more trials, I had it performing much better -- still a touch slow, but not agonizingly slow compared to the other cars. [Regretfully, I didn't write down what I did, so I can't pass on the tip... likely I lengthened the late apex so the AI could keep the tails in line when they got on the gas or smoothed the braking zones so they weren't pitching the car sideways on corner entry]
And this is the biggest point with the AIW: you only have one AIW for each track, yet it has to work for all cars, so it is extremely unlikely to be perfect. If a better AIW is needed for a particular mod, you're probably better off training the AI for that specific mod rather than trying to fix your AIW as long as the mod is only 1 out of 20. I don't drive formula cars very well, so I perfect my AIW in AE86, NAGT, & Historix. When I'm satisfied with them, I mostly watch the AIW in other mods to see that the AI are not doing something evil and trust that the performance is where it should be (e.g. if I switch to F3, then I better not be quicker than the AI at 100%!).