Barton,
Another quick response....
There are known bugs in integer-decode-float in GCL 2.5.0. These have
been corrected (as Camm reports) in more recent GCLs. I would recommend
you use a current GCL if you want to play with floats.
As for NaNs, Maxima currently does not support any of the non-standard
floats (NaN, Infinity, etc.), and I don't quite understand what GCL's
story is. In any case, introducing them blindly into Maxima will break
things in the simplifier. Compare also bug report 781726 (e.g. 2.0^1024
=> internal error; is(2.0^1024=2.0^1025)=> TRUE, etc.).
I suggest you review the mailing list archives and the bug list, because
I think some of your other points have also already been reported.
-s