There appears to be something wrong with your question. In general variation within a group are expected to be less than the variations across the group. Your question states reverse of this.
Anyhow, when a population is composed of identifiable groups, there are two broad sampling approaches possible. In the first approach, the sample of required size is drawn at random from the total population. In the second approach, the size of sample to be drawn from each group within the population is fixed. The size of the sample is generally related to expected variation within the group. But, when variations within the group are not known, or when it is desired to use a simple method of analysis, the sample size from each group may be proportional to the group size.