Earlier quoted context omitted.
Smaller effect size as studies improve is the classic sign of a non-event. ESP and the like went through similar cycles - impressive early studies followed by a series of more rigorous studies with diminishing effects until people got bored and moved on to the next big thing.
In this case different tests have occurred at different power levels. Has anyone normalized the results? Of course with only a couple replications its early for a meta study.
I think that the best metric is to compare this with the maximal Force/Power ratio that is consistent with a "massless" thruster and special relativity, (i.e. a laser beam, photon thruster, ...). In this case the Force/Power ratio is at most 1/c, that is 0.0033 uN/W = 0.0033 mN/KW.
The results in the table vary from 3x to 200000x the theoretical maximum if we assume that this doesn't break the conservation of momentum, special relativity and quantum mechanics.