obi clean behaves unpredictably on huge datasets, sometimes using a lot of memory or spending a lot of computation time in the tsearch library. Replacing the tsearch library with my AVL tree implementation for example could fix those issues.