The number of observed runs. Larger samples reduce variance of the estimate and tighten confidence intervals.
Comparing two routes with similar EVs requires enough runs to overcome randomness.
Small samples can be misleading, especially with rare, high-value drops.