I have changed this value up to 100001 which is the maximum value choosable. My results were completely different changing this value, but I was simulating a resonant structure with very thin resonances so I needed a high frequency precision. Anyway, in my opinion you should always do a test of convergence of the results.
well i have done some simulations by parametrizing number of frequency samples .
i haven't observed any noticeable change in the results.
so i can conclude that this parameter has not a significant affect on results.
however as my models were mostly antenna related so this conclusion may be valid only for antennas .