The reason sigma clipping normally uses multiple iterations is because the sigma₁ estimate in the first iteration includes the outliers, and is thus biased to have a larger value. Once these outliers are removed, the sigma₂ estimate in the next iteration will be smaller, and thus identify more outliers outside the (now tighter) bounds. This is why you get progressively fewer points within the bounds, because the sigma estimation keeps ignoring more and more outliers away from the mean.