That's some... peculiar dispersion. Looks like on axis response slightly compensates for it but I can't imagine optimum linearity holds up over a particularly wide angle. Clearly these should be used as actual nearfields in a well-treated environment.