I personally am starting to think there's something to Amphion's approach of emphasizing the region from 2-5k, which seems to tie in quite closely with my own tastes.
Though the speakers were centered for the preference studies, from what I can tell not all the research is conducted with centered speakers. The most important thing seems to be consistent coupling with room modes.
In his book, Toole describes one of Harman's test rooms -- maybe the same one Amir was in -- mentioning that the programmable speaker mover "can compare up to four single loudspeakers at one time, up to four stereo pairs, or up to three L, C, R combinations. The single loudspeakers can be compared at L, C, or R locations at varied distances from the side walls. The distance from the front wall is also a controlled variable." See figure 3.13.
In chapter 7.4.2, he talks about one of his studies on directivity, (the one about narrow vs wide and stereo vs mono) and the illustration shown has the speakers at the corners.