I found an issue in cp that caused 350% extra mem usage for the original bug reporter, which fixing would have kept his working set at least within RAM.

http://lists.gnu.org/archive/html/coreutils/2014-09/msg00014.html