Hello again. I've been otherwise engaged for a while and am catching up. You are right that the strongest evidence linked to perceived sound quality in a room is a comprehensive set of anechoic data. Room curves are only loosely connected, except at low frequencies where they are the only reliable data. When we see a family of on- and off-axis anechoic curves, all smooth and indicating either constant DI or gradually rising DI we have learned (a) that there are no audible resonances and (b) that the spectral balance of direct and reflected sounds will have something in common. Which is more important? I vote for the absence of resonances, because timbral colorations of that kind are easily heard and corrupt all program material independent of the listening space. Neutral loudspeakers tend to "disappear" behind the double-blind screen, allowing the spatial cues in recordings to be heard.
The matching of direct and reflected spectral balances is less well defined. In Figure 7.12 in my current book describes an elaborate test done around 1985. I show that the least preferred loudspeaker had arguably the most constant DI, but was more directional than the higher rated speakers with quite uneven off axis radiation, but more of it. Listeners seemed to have voted for more "space" and seemed to be able to rationalize the low-Q directional variations. BTW the spatial perceptions cannot be "envelopment" in the classic sense because that is correlated with much delayed (e.g. 100 ms) sounds, as in concert halls. This is why multichannel reproduction has huge advantages. What is it then, that causes the preference for some reflected sounds? My personal theory is that it has to do with the fundamental limitation of two-channel stereo, in which some of the sound almost inevitably emerges from the L or R loudspeakers - hard panned by console controls or by mic techniques. When this happens close miked solo instruments or groups of orchestra instruments emerge from a point in space. I have found this to be highly annoying in some recordings as it is plainly unrealistic. Spraying the sound around adjacent walls softens the mono L&R images. It also softens the amplitude panned images between the loudspeakers which some people disparage and others find appealing. The musical genre is likely a factor. But the basic issue is that conventional stereo cannot deliver the goods and we look for ways to improve it.
Is constant directivity, including omnidirectIonality, better than the slightly increasing DI of conventional forward-firing cones and domes? Having heard both over many years I have no preference. If there are no audible resonances the difference devolves to the localized sense of space and delocalization I just described. Neither is more "correct" in stereo reproduction. In multichannel systems it simply does not matter - with more real sound sources direct sounds dominate. Forward firing designs occupy less room volume and are likely to be preferred.
My current preference for multichannel upmixing is based on its superior performance and its adjustability for different programs. Such a system also permits playback of multichannel music recordings, music videos (some of which are excellent) and movies. If your circumstances permit it, I highly recommend it.
Here is a link to a description of my present system:
https://www.dropbox.com/s/gg5wnct6010gvuq/3-Part Toole Home Theater 2018.pdf?dl=0