As Slawek observed, expanding runs of the value zero in RLE encoded coverage data is both pointless and expensive. A simple optimisation is to avoid all together by checking the value of the run before expanding.