We have seen a few cases of observations of bright sources using Suzaku where running cleansis creates a hole in the image. Part of the problem is running with too high a log probability parameter but there is also a subtle error when iterating. Pixels which are already identified as bad and have their counts set to zero are still included when calculating the local background. This artificially lowers the estimated background hence increases the probability that a pixel will be flagged as bad.
I changed the algorithm so that pixels flagged as bad in previous iterations are ignored when calculating the local background. This change has been checked into the development version.