For example, in my direct experience (not with phones, but networking gear), the way the product is placed in the test fixture and the way cables are dressed has an impact on emissions and measurements. In one scenario, as a product engineer supporting equipment through a certification lab, I had the mandate to pass with a 3 dB margin from the company I worked for, but, after days of testing and failing, it was decided that the equipment under test could pass with cables routed in a special way (which was not uncommon for normal customer use) and with quasi-peak detection (i.e. averaging the emission peaks) that ended up below the regulatory (FCC for example) passing below the regulatory mask/level by a fraction of a dB. I have always wondered whether a different test lab, without support from product engineers, would pass the same test. I think the answer in "no".
Edit: Quasi-peaking was not a dirty trick that was pulled to pass certification. It was an acceptable method of measuring emission power suggested by the lab technicians.