I doesn't matter. When they are talking about eigenfrequencies, they mean the frequencies of the modes, resonant modes, of the structure. It may not necessarily accord to the frequencies where the permeability and/or permittivity are negative. Of course it shouldn't since the resonance means a change in sign in the real/imaginary part and a peak in the imaginary/real part of the parameter (this can be verified via the Kramers-Kronig relation). So at the resonance the permittivity or permeability is generally ill-defined (in an ideal resonance, in reality it's just going to be very small) since it is changing sign. We do not want to operate at the resonance mostly because it corresponds to a peak in the loss (imaginary part). What you generally do is design the SRR so that the resonance is a bit off from the operating frequency so that you can not only be in the small area where the parameter is negative but also try to compromise on the amount of loss.
EDIT: Normally though we do not consider the surrounding dielectric (if there is one) to be nonlinear. As pipe-squeezed-angle_bracket explained, the resonances of the SRR are dependent upon the incident field. If we permeate a volume or surface with our SRR and consider the bulk behavior of this volume/surface so that it has an effective permeability/permittivity then we find that it is non-linear (for the reasons why the physics of the SRR are dependent upon the incident wave).