I think, what you are getting is not a true representation of dielectric constant. As you are having some metal particles in your sample, they are causing a leakage current in your sample. This is giving results to (1) very high apparent dielectric constnt and (2) a unstable measurement values. I have numerous sample measured at various voltages (k vs. V). Near the breakdown voltage where there is a considerable leakage current in the sample, near DC measurements (even upto 1 MHz) can show very high k values as well as negative values. However, there results are not reproducible. In that low frequency regime, I think, L and R plays a dominant role for conductive samples instead of C. However if you are measuring at microwave freq, low skin depth in metal may prevent such thing to happen. However you will see a high loss tangent in the sample.
This may be opinion from experience and your case mey be completely different.
Dear Raju, looking at your dielectric constant results, it will be viable if you try to understand the dielectric behavior with tandelta to understand what B Roy explained above. You should start with pure ferrite and add very small amount of %metal for some real good explanation. As increase in metal %age and the thermally activated carriers with temperature will complicate dielectic phenomenon explanation.
I agree with the colleagues above. It might be helpful if you describe the experimental arrangement (size, shape of the sample holder, nature of your sample, volume fraction of the sample in resonator, applied voltage etc.). 100 kHz is not a specially hight frequency. Did you try to use compression molded parallel-plate geometry with standard AC bridge method? First you should try to do isotheraml measurements at different frequencies to the stability of your results.