Since there’s no expectation of that uncertainty being systematic, taking a relative change pushes your error bars +3dB.
That 6dB drop is under the error bars, so it’s pretty meaningless, and that’s with a professional setup.
In other words, there’s no expectation that the numbers you’ll get from these will translate into anything useful, at most you have a ‘hotspot’ detector.
This kind of debugging is very routine. You do it manually with E/H field probes and a spectrum analyzer in between lab visits. This setup makes it more repeatable by automating it.
To comply with CISPR (as an example), a lab only has to validate <4.5dB (in the 30 MHz to 300 MHz range), and up to 5.2 dB (from 30 MHz to 1 GHz). See for instance [1].
By the way, if you are doing this kind of testing routinely, you need to check your PCB layout process. This fine grained tweeting should be a last resort option.
[1]http://ing.univaq.it/emc-chap-it/download/MUinEMC_Revision%2...