I was fairly certain that the zero-taper-top was a big part of my screwup (when all else fails, FOLLOW THE TEMPLATE!) so I cut that off completely, and left the sides as is. MPG immediately jumped up from 13.7 to 14.4 and it was noticeably easier to cruise on the highway.
I moved the 3rd brake light to below the windows, so I can do a variety of experiments and not have to remount the lights all the time. It's legal, since they only need to be roughly centered and above the height of the left and right tail lights. As expected, the tuft testing on the side looked great, with no inversion, no separation, and very little visible turbulence. It may not be the optimal departure angle (especially near the top), but the tuft testing showed that it was at least not a disaster zone!!!
I added back in the center section, this time with a slight taper to it, about 3 inches down over 20 inches run. That's about an 8-10 degree departure angle. The corners are still wrong, but my MPG stayed consistent at 14.3-14.4 for 2 tanks. This isn't really an improvement over no center section, but it's a lot better than with the old moronic one! It's also pretty nice looking on the tufts. However, I'm looking for improvements, so I made myself a wooden taper gauge. I'm currently experimenting with a 12 degree departure angle and a 1.5 degree per inch taper rate. This is steeper than The Template, but it matches what Mercedes and Ford have done with the Sprinter and Transit vans. They both have fairly flat sides, and then go to a 10-12 degree taper at the back end. Ford uses a pretty sharp kink to get to that angle, and Mercedes uses a smooth curve. Here's my partially-carved build of foam to get there, you can see my taper gauge on the left:
