Embedded hard real-time systems that are based on software product lines using dynamically derivable variants are prone to overestimations in static WCET analyses. This is due to the fact that infeasible paths in the code resulting from infeasible variant combinations are unknown to the analysis. This paper presents an approach to incorporate variant constraints in the calculation to exclude infeasible paths and thus to decrease the WCET overestimation. Based on feature models the authors propose a sound approach to identify significant infeasible paths that can be safely discarded in the analysis. The benefits of the approach are exemplified by a real world example from the automotive domain where they are able to reduce the WCET bound by up to 50 percent.