That article is a little light on the details of the test setup and measurement procedure but, if I'm reading it correctly, he played pink noise at a fairly high level for ~90 hours straight. The measured differences are likely due to voice coil heating and slight position and seal variations over time.