But I had understood it wrong: I thought that they tested three devices, but in reality they only tested a Cannae and the Null article and BOTH produced thrust.
Incorrect. There were 3 test articles. 2 test articles were provided by Cannae. The articles were the same, except that one of the had holes on one side the RF cavity. The third article was a 50 ohm resistor. Only the third article produced no thrust.
The test article is a short, wide cylinder. During the test, the device is mounted horizontally ('cause gravity). Fetta (Cannae) theorized that the device works by having particles bouncing off between the "left" and "right" walls of the cylinder back and forth. However, in such case the net force is zero, because the force exerted on the left plate is the same as the force exerted on the right plate. (This is test article #2, a.k.a. "unslotted"). So he had another article, which had holes drilled in the "right" plate, permitting particles to escape. If the holes were, say, 20% of the plate surface, then the force on one plate would be reduced by 20% -- as 20% of particiles would escape through the holes -- and this would produce thrust. (This is test article #1, a.k.a. "slotted").
Test article #1 produced 40.0uN of thrust on average, test article #2 produced 40.7uN of thrust on average. Since there's some spread in the data from run to run, it looks like both articles produced the same thrust.
Test article #3 (a.k.a. resistor) produced no thrust. (This is not entirely true -- the authors say that they have measured 9uN of force in the configuration involving the resistor due to magnetic field from the feed cable pushing the torsion pendulum, so they have substracted 9uN from all results.)
Most of the confusion stems from the fact that the abstract refers to the "unslotted" article as null. This is because it was a null for Fetta's theory that the thrust is produced by holes in the wall of the RF cavity.