I have experimented quite a bit with matching subs to my LaScalas, and have personally found that the horn loaded subs sound best by a wide margin. I was never sure exactly why, but I was just reading through a very long thread where Lynn Olson mentioned something that gave me an epiphany. Here is the snippit:
Lynn Olson - diyaudio thread top of page 3 pgph 7
I've been thinking quite a lot about dispersion characteristics - most horn-fans don't know it, but unless they've got an all-horn system, the dispersion is not in fact constant with frequency. The direct-radiation bass unit radiates over 360 degrees (omnidirectional), then gradually narrows down to 180 to 90 degrees (depending on crossover point), then hands off to a horn with the specified radiation pattern (typically 120 to 90 degrees), which then narrows further as the frequency increases (unless it's a constant-directivity horn, which have their own problems). The narrow sidelobes that appear in the polar-pattern curves also appear as ripples in the time-domain, frequency-domain, and impedance curves - this is a consequence of antenna theory, where errors in one domain must appear in the other domains as well.
Lynn is one of those people who can explain things well. I am not going to try to expand on this, but I will say that it makes PERFECT sense. To my ears, a radiator sub is a mismatch. My brain is recognizing the errors in the time-domain and frequency domain, and even though I don't really get why, it just doesn't sound right.
Thank you Lynn. I guess there is a lot our brains and ears do that we are really not cognitive of.