The evaluation of the energy-efficiency of applications and protocols is one of the most important issues in Wireless Sensor Networks (WSN). However, this is a time consuming and challenging task. Therefore, a realistic energy-efficiency evaluation is often neglected or oversimplified by using simple theoretical models or unsuited simulation tools. In this work, we evaluate the energy-efficiency of two specific key exchange protocols, an Elliptic Curve Diffie-Hellman with authentication (ECDH-ECDSA) and a Kerberos based approach, while using different duty-cycling MAC layer protocols. In our evaluation, we compare the results from a theoretical model with simulation results using the AVRORA+ simulation tool and real-world measurements in our WSN testbed SANDbed. Using three MAC layer protocols, Tiny-OS built in Low-Power-Listening as well as S-MAC and standard 802.15.4, we show that there are several important cross-layer effects that should be considered when performing an energy-efficiency evaluation. Furthermore, we argue, that there are several evaluation metrics like the absolute energy consumption per key exchange or one key exchange per measured time interval.