Heres another one of many comments regarding gcc 2.7.2:

I'm experiencing evil floating point problems on my 1.1-current system
which has been compiled with gcc 2.7.2.  These problems don't appear on
a stock 1.1 machine, and don't appear with a static binary built on the
1.1 machine.

I'm presently trying to track down the culprit by rebuilding the 
libraries with gcc 2.4.5 -- however, I did get up at 9am so the chances
of operator error are high -- if you'd like to look at this code, its:

ftp://resort.com/pub/lucas.c

compile and run with "a.out 17 32 0 | more", if you see a lot of NaN in
the second page, then you've got the bug too.  It appears that between
the end and beginning of a while loop, that the value of a double goes 
from 0.0000(small) to NaN.

-j