Step 1: Recall the ideal structure for clusters. In cluster sampling, an entire cluster is selected and every unit inside it is surveyed, so each selected cluster should behave like a small mirror image of the whole population.
Step 2: This means the units within a cluster should be as heterogeneous (varied) as possible, so the cluster captures the full range of the population's characteristics. If a cluster instead has very similar (homogeneous) units, the information from repeatedly surveying similar units adds little extra precision, wasting sample effort.
Step 3: This is the exact opposite requirement of stratified sampling, where strata are made internally homogeneous. For cluster sampling, high within-cluster variation and low between-cluster variation is what makes the design efficient, since fewer clusters are then needed to represent the whole population well.
Final Answer: Cluster sampling becomes more efficient when clusters have more variation among the units within a cluster.