One thing about the saw tooth that always confused me is that it seems like a leading edge sawtooth is creating a 2D corner box effect. Like the reflectors on the side of your car. So there would be a narrow angle, directly in front of the sawtooth where the return should be huge. I'm guessing that there's something I'm overlooking.
Only if it's a 90 degree angle. A sawtooth for stealthing is going to be at, perhaps, a 50 or 55 degree angle - it won't bounce back to the original source.
Basically if you draw any angle onto a subject, it'll bounce off at the same angle opposite the perpendicular; with a 90 degree angle this means that your radar beam will bounce twice and come right back at the source. Having differing angles in the sawtooth bounces radar in other directions, it doesn't matter too much between higher and lower than 45 degrees as long as it's a significant enough amount. See case B and C for that; the decision about what angles to use will be much more dependent on other aspects of the aircraft. Things like the angles made by the YF-23's wings, for example; they're all the same so that you only get a hard reflection in one direction instead of several (think of the front edge of each wing as a flat plane reflecting radar energy) and so you'd want to match that angle with your other little details. That way you only get one "lucky" angle to be at to have significant detection capability, and the further that is from a nose-on aspect the better - it means the only people that have a chance to detect you are the ones farthest from your target.
Anyway, just showing angles here. Yes, they're all 90, above or below, but 45 degrees from what the surface would be.