In our usecase, these are impossibly greedy because we often free memory in a different thread than where we allocate, forcing the default allocator to scatter the used memory all over the place.