Improving the evaluation of samplers on multi-modal targets