I'm designing a two-way loudspeaker system using Dynaudio drivers. I
modeled the impedance and acoustic frequency responses of the drivers
using CALSOD from measurements that I've taken. I designed the
crossover for each driver in order to get 4th order L-R alignments. So
far so good. My question is, since drivers aren't perfect, should I
optimize the final -system- response for flattest frequency response
or should I try to get as close as possible to the theoretical L-R
alignments? If I try to get flattest system response then each driver
+ crossover section deviates quite a bit from the theoretical L-R
alignments. I may get very flat frequency response but I lose
advantages of the L-R alignments.

Which is the better compromise for getting best subjective results?

Thanks in advance,

Michel