We address the issue of sharing the CPU and RAM resources of a cloud computing system. Specifically, we argue that these resources should be shared according to proportional fairness instead of dominant resource fairness. The latter has recently been proposed as a desirable scheduling objective mainly because of its strategy-proofness property. We show that proportional fairness, which satisfies a weaker form of strategy-proofness that we refer to as scale-invariance, yields a much better resource utilization and outperforms dominant resource fairness in the realistic setting where jobs arrive and leave at random times.