Mysterious test_pareto failure on Travis

See here for a discussion:

http://mail.scipy.org/pipermail/numpy-discussion/2012-September/063801.html

The failure is caused by Travis switching to 32bit Ubuntu 12.04 recently. Some change in Ubuntu breaks it. Even this old commit 3882d65c42acf6d5fff8cc9b3f410bb3e49c8af8 still has the failure, so it's probably gcc related. The test is probably not robust enough.
