The idea is to attempt to correct for the nonlinearities in the amplifier, not necessarily linearize the output (into a reactive load). As such, you really don't want a highly reactive load at the feedback node...I would guess you'd want a resistance of similar impedance, or a less reactive simulation of your target load.
Whether or not this is a good idea is another thing entirely. To me it seems as though this really is a no feedback design that happens to be extremely linear (into a resistive load). Most low-distortion, no feedback designs I am familiar with are stages driving high-impedance resistive loads--not the case with loudspeakers. I think you'd have to build it to really be able to say much about it as a concept.