By the way, you're wrong on both points - the cache of page mallocs is not infinite, and it does spill dirty pages to disk when necessary. And the latter is what bounds the number of malloc'd pages.