It is impossible to generate momentum in a closed object without emitting something, so the drive is either not generating thrust, or throwing something backwards. There is no doubt about this.
Assuming that the thrust measurement is accurate, that something could be radiation. This explanation is exceedingly unlikely, since to get mN of radiation pressure you need an enormous amount of energy, since in 1s you get 1 gm/s of momentum, which in radiation can only be carried by 3*10^5 J (multiply by c), so you need 30,000 Watts of energy to push with mN force, or at least a million Watts for 80mN. So it's not radiation.
But a leaky microwave cavity can heat the water-vapor in the air around the object, and the heat can lead to a current of air away from the object. With a air current, you can produce mN thrusts from a relatively small amount of energy, and with a barely noticible breeze. To get mN force, you need to accelerate 300 cm^3 of air (1 gram) to 1 m/s every second, or to get 80mN, accelerate 1 m^3 of air (3000 g) to .2 m/s (barely perceptible) and this can be done with a hot-cold thermal gradient behind the device which is hard to notice. If the thrust measurements are not in error, this is the certain cause.
So at best, Shawyer has invented a very inefficient and expensive fan.
EDIT: The initial tests were at atmospheric pressure. To test the fan hypothesis, an easy way is to vary the pressure, another easy way is to put dust in the air to see the air-currents. The experimenters didn't do any of this (or at least didn't publish it if they did), instead, they ran the device inside a vacuum chamber but at ambient pressure after putting it through a vacuum cycle to simulate space. This is not a vacuum test, but it can mislead one on a first read.
In response to criticism of this faux-vacuum test, they did a second test in a real vacuum. This time, they used a torsion pendulum to find a teeny-tiny thrust of no relation to the first purported thrust. The second run in vacuum has completely different effects, possibly due to interactions between charge building up on the device and metallic components of the torsion pendulum, possibly due to deliberate misreporting by these folks, who didn't bother to explain what was going on in the first experiments they hyped up. Since they didn't bother to do a any systematic analysis of the effect on the first run, to vary air-pressure, look at air flows with dust, whatever, or if they did this they didn't bother to admit their initial error, this is not particularly honest experimental work, and there's not much point in talking about it any more. These folks are simply wasting people's time.