Returning to the classical diversity-stability problem about population densities, we propose a hypothesis that the spatial sparseness of food plants of herbivores, which is frequently observed in diversified communities such as those in the tropics, may be a potential source of the low variability in herbivore densities. The hypothesis consists of the following components: (1) sparseness of food plants of several herbivores increases with increasing number of plant species in a given amount of area, (2) spatial sparseness of food plants reduces the dispersal rate of herbivores between food plants, (3) such a decrease in dispersal rate causes a larger spatial variance in the consumption rate of food plants, (4) such a large spatial variance in the consumption rate of food plants reduces the temporal variance in the amount of food plants, and (5) the reduced temporal variance in the amount of food plants yields a smaller variance in the temporal dynamics of herbivores. A simple mathematical model was constructed to show a theoretical basis of the fourth component of the hypothesis. To exemplify the second, third, and fourth components of the hypothesis in fields, we conducted field experiments on the interaction between butterfly larvae and cabbages, in which cabbage seedlings were planted by two levels of sparseness: 1 m apart and 0.2 m apart. The spatial variance in the amount of leaf dry weight that escaped from larval feeding was larger in sparsely planted fields, which was in accordance with the third component of the hypothesis. The variability in the mean amount of leaf dry weight was smaller in sparsely planted fields, which was in accordance with the fourth component of the hypothesis.