Recent observations made by Kanamori and Allen about earthquake recurrence time and average stress drop revealed a very interesting relation: earthquakes with longer recurrence times have higher average stress drops. They attributed the difference in stress drop to the difference in long-term average slip rate. To interpret their result in terms of the healing effect, we simulated earthquake recurrence with a one-dimensional mass-spring model, incorporating a recently developed rate-and-state dependent friction law for different loading rates and heterogeneous strength distributions. We first calculated the stress drop and recurrence time as functions of loading rate for a homogenous fault model. We found that the stress drop increases up to 30% when the loading rate decreases from 10 cm/yr to 0.01 mm/yr. Thus, the observed great variability of stress drop, from a few bars to a few hundred bars, which is obtained by replotting the data of Kanamori and Allen in the form of stress drop versus long-term slip rate, may not be attributable to the healing effect alone. Our numerical simulation shows that the variability may be due primarily to the spatial heterogeneity of strength on the fault. Our simulation also suggests that of the two empirical laws that were inferred from the same laboratory friction data, called the power law and the logarithmic law by Shimamoto and Logan, the former can explain the observed relation between stress drop and slip rate better than can the latter, at least for strike-slip fault. The logarithmic law is an earlier and simpler version of the rate-state-dependent friction law.