One important point discussed between practitioners in insurance business is how many Monte-Carlo simulations are necessary to estimate the required Solvency Capital. A widespread opinion is that the larger the size of the Monte-Carlo sample is, the more stability of the final estimate can be achieved. By analysing the properties of the risk estimator from the point of view of Robust Statistics, we argue that augmenting the sample size in Monte-Carlo simulations is of limited benefit for the estimator's stability. In particular, the local-shift sensitivity is seen as the term that accurately reflects the problems arising with the numerical issues in Monte-Carlo simulations. A semi-parametric approach built on generalized Pareto distributions is applied to the tails yielding stable estimates.