In this paper we challenge the common evaluation practices used in Virtual Machine (VM) consolidation, such as simulation and small testbeds, which fail to capture the fundamental trade-off between energy consumption and performance. We identify a number of over-simplifying assumptions which are typically made about the energy consumption and performance characteristics of modern networked systems. In response, we describe how more accurate models for data-center systems can be designed and used in order to create an evaluation framework that allows the exploration of the energy-performance trade-off in VM consolidation strategies with enhanced fidelity.