I suspect this is a result of matrix decoding steering data from the mains to the center and surrounds (in other words, a feature). Is this occuring when listening to the same source material with the different processing modes applied?