Considering you somehow coerced a friend into stop watch duty, it shouldn't be too hard to tuft test your kammback. This will tell you if the angle is right- and perhaps allow you more improvement.
I would also encourage you to tuft test the rear of the car without any aero aids. If you can figure out where the air starts to get turbulent behind the C Pillar, you can find where to end the NASCAR spoiler. If the air is clean and attached, no spoiler needed. You just want to trap a vortex behind your turbulent window to act as a virtual kammback.
Should this work, you can use a piece of stiff, clear acrylic instead of coroplast. This will allow you full vision while taking advantage of the aero benefits and Coventry data on base plates.
I do not know how big or in what shape you should build a spoiler. You might, however, be able to rig up a test apparatus to find out. Instead of tuft testing only on the surface, you could try another technique the professionals use- 3D tuft testing.
Looking through some old Car Styling issues, I found these photos. Issue 2009.1, page 16.
From a square framework, they strung cables vertically behind the vehicle. From the cables are attached tufts about 5cm apart from each other in a careful array. You can see that the ones going straight back are "good" flow and those that are fluttering exhibit turbulence. By attaching a frame like this to the top of your trunk you
may be able to see where to build a spoiler, or even kammback.
I don't know if this will work in real world situations or if it's purely a closed-environment test. Maybe aerohead can weigh in on this; it might be a fun thing to try.
PS- I'm also interested in seeing the raw data. 3-4% isn't much, and if the data varies a bit those kind of changes may not be conclusive.