Prof, you're right about the "hollowing effect". Phase cancellation is a central problem making first-order implementation very treacherous. The output of 2 combined drivers produces a lobe which cancels or augments output when the "listener" is not on the proper vertical axis to integrate the two signals correctly. The ideal way to implement phase coherence requires a true point source, as in the coincident upper drivers of more recent Thiel designs. Short of that coincident solution, the closer the driver spacing the better; notice the touching/clipped perimeters of Thiel mid-tweeters. Plus, the crossover frequency must be as low as possible because the beaming of the upper end of the larger (lower) driver interferes with and is discontinuous with the radiation pattern of the lower end of the upper (higher) driver, making for a room power response different from the on-axis driver response. To wit: Thiel upper crossovers are very low, lower than considered feasible by most; Thiel tweeters cross in around 3KHz. Such a bold XO frequency requires a tweeter that behaves all the way down, below 100 Hz. At these low frequencies there are bound to be resonances which would destroy the tweeter if not mitigated with notch filters, which are themselves costly. Broadband drivers are a design feat as well as a management challenge. Most experts, including those at the "New Thiel" deem the task "impossible".
You are correct: the coincident drivers do a better job than the older multi-driver solutions. But those multi-drivers are themselves extremely sophisticated and allow very low crossover points and both physical and electronic resonance control. Part of Thiel's low impedance (which we love to hate) is that each resonance correction lowers the system impedance, and the global system is really not serviceable until all resonances are effectively eliminated.
As I perform my XO upgrade investigations, I am continually surprised how good all these Thiel drivers are. As an example, the non-resonant bandwidth of the 3.6 midrange spans 7 octaves, with similarly stellar performance from every driver in the stable. Part of my decision to reach back no farther than the CS2 2, is that previous drivers were merely modifications of off-the-shelf units from Dynaudio, Seas, and so forth. Newer drivers incorporate new sophisticated technologies toward success in first-order systems.
Back to the lab.