I was suggesting using VIX and SP500 (and similar VIX like index) to backtest, and not to worry about strike specific vol. Of course this is just an approximation.
Oh, I see. You would be better off just using ATM vol and doing simple RV-IV tests then. Variance swap includes a fair amount of convexity premium, so it will always come in as expensive on RV/IV basis since the var swap payoff = (realized_vol^2 - var_strike^2)/(2*var_strike).