I agree that the portable C build is less than ideal - but does it create a significant performance hit wrt the way it's being used by Net::SSH::Perl ?
The "manual procedure" still works fine for me with pari-2.1.7 and Math::Pari-2.010703 on both Win32 and Linux - but that's on an ix86 architecture. Check out
http://rt.cpan.org/Public/Dist/Display.html?Name=Math-Pari and see if there's anything there that helps. In particular, click on the link to bug 12324. Seems that someone made some progress on sparcv9 with the previous version of Math::Pari.
Cheers,
Rob