Within classical MAPE-K control-loop structures for adaptive systems, knowledge gathered from monitoring the system and its environment is used to guide adaptation decisions at runtime. There are several approaches to enrich this knowledge base to improve the planning of adaptations. We consider a method where probabilistic model checking (PMC) is used at design time to compute results for various short-term objectives, such as the expected energy consumption, expected throughput, or probability of success. The variety PMC-results yield the basis for defining decision policies (PMC-based strategies) that operate at runtime and serve as heuristics to optimize for a given long-term objective. The main goal is to apply a robust decision making method that can deal with different kinds of uncertainty at runtime. In this paper, we thoroughly examine, quantify, and evaluate the potential of this approach with the help of an experimental study on an adaptive hardware platform, where the global objective addresses the trade-off between energy consumption and performance. The focus of this study is on the robustness of PMC-based strategies and their ability to dynamically manage situations, where the system at runtime operates under conditions that deviate from the (idealized) assumptions made in the preceding offline analysis.