Codes based on branch and bound techniques seem to have problems with
some of the test problems from the Mor'e et al. test set.
For example,
Peter Spellucci <spellucci@mathematik.th-darmstadt.de> reports
for Knueppel's PROFIL/BIAS Interval Branch and Bound code CPU times of
16 minutes for the Woods funktion (n=4), and memory overflow
(72MB user) for Kowalik-Osborne (n=4) and Bard (n=3). The reason seems
to be a pessimistic natural interval extension that makes the
bounding part extremely weak.