As in any speaker, you want to keep the width of the driver small compared to the wavelength of the highest frequency it handles, taking into account the slope of the low-pass filter on the crossover. The height of a line source is ideally floor-ceiling as Wendell said; it behaves like a much larger line source if you do that, several times the height of the room. If it's shorter, you'll get vertical lobing at lower frequencies, and as Davey just pointed out the radiation pattern will change with frequency and listening distance. The flip side of this is that the smaller the diaphragm area, the larger the excursion needed for a given SPL. So you want to make the drivers as wide as possible without making them so wide that they become directional. With cones, planars, and ribbons, you can make a pretty good compromise, but with stats, you have to use other techniques such as facets (Sound Labs), curves (Martin-Logan), and phased arrays (Quad) since they can manage only about 1/10 the excursion of a planar. (All three techniques do essentially the same thing acoustically.)
The MMG would be an example of a Maggie with compromises -- the quasi ribbon tweeter is a bit too wide, something like 2", so you get beaming and cauliflower lobing in the horizontal dispersion pattern, and the speaker is too short, so you get vertical lobing as the driver and its reflected images cause interference patterns. The 20.1 with its floor-ceiling height and 1/4" tweeter and narrow midrange (sorry, I don't know the actual dimension either) is an example of a cost-no-object design. In general, the more frequency ranges you have, the more closely you can approach the twin ideals of large diaphragm area at lower frequencies (since excursion increases as the square as frequency drops, as the cube at the lowest reaches of a dipole) and narrow width. But of course adding drivers increases cost and size, and you have the adverse sonic effects of more crossovers.
