Strange behaviour

[Lonnie Princehouse]

> Any idea about the origin of this problem?

My setup is identical except with Xeons instead of Opterons, and I get
the expected result with Numeric 23.1.  This indicates that it's
either a 64-bit bug with Numeric 23.1 (which looks like it's been
fixed in your latest-cvs test), or the ebuild is compiling wrong.  The
latter seems unlikely, but building 23.1 from the tarball instead of
emerging it, and then running your test again, would be decisive.