I haven't tried to run it, but I just looked at
loading maxima-lapack and running your eigensys program.
It would need to be bfloat-ed if the objective is to use it for
arbitrary precision rootfinding, using the companion-matrix idea.
A more interesting task might be to make a modified f2cl that generates
bfloat code.
Or see this:
"Yozo Hida sent a pointer to his Perl script around. The script
transforms double and double-complex routines into single, quad,
and other precisions."
In the lapack meeting notes from March, 2006.
http://icl.cs.utk.edu/lapack-forum/archives/lapack/msg00039.html
Yozo is a student at Berkeley... I'll try to get more info from him :)
RJF