One of the approaches to compare forecasts is to test whether the loss from a benchmark prediction is smaller than the others. The test can be embedded into the general problem of testing functional inequalities using a one-sided Kolmogorov-Smirnov functional. This paper shows that such a test generally suffers from unstable power properties, meaning that the asymptotic power against certain local alternatives can be much smaller than the size. This paper proposes a general method to robustify the power properties. This method can also be applied to testing inequalities such as stochastic dominance and moment inequalities. Simulation studies demonstrate that tests based on this papers approach perform quite well relative to the existing methods.