My comments and conclusion are based solely on this particular model. The changes seen will likely differ for another system with a different impedance curve. One must also remember that this was for a purely resistive load, one which was exagerated when compared to decent cable. If you use 18 gauge zip-cord, well, you may have worse conditions than those presented here. Especially if it is a long run.I conclude that the impedance compensation is best when located at the speaker crossover. Even when the compensation is applied only to one crossover section (the case here), the sensitivity to cable impedances is less than if it is connected at the amplifier terminals.
Even though the amplifier will see an impedance which is not perfectly resistive (due to the cable impedance) the variation in output from the speakers is what is paramount. This set of graphs shows what will happen for a perfect amplifier load. What you'll actually have will be more varied, especially if you use tube amplifiers or cheap, long runs of zip-cord.
Prior to this evaluation, I had dismissed using zobels since the CAD software could produce an excellent crossover without them. This did not take into consideration differing cables or transients.
Transients are not considered in an optimization such as this. They are more complex and I do not have the means to evaluate them. However, anything which will reduce variations related to cable impedance should be considered beneficial, especially if you want or need to use long runs of cable.