Just some background to this thread....
The standard mechanism at FNAL for distributing the Kai C++ compiler is
through KITS using upd. There are some minor local patches made to the Kai
product here at FNAL, such as adaptation to ups/upd, fixes to some system
header files inside the Kai package (for Linux2 and IRIX6 in particular), and
combining different ABI variants of Kai C++ into one package (on IRIX6). You
will not get these changes in Kai's RPM, though they do sometimes react to our
modification requests by adapting headers in later patch releases.
Hence, "officially", the problem reported would have only been encountered by
the FNAL packager of Kai C++ (was myself, is now Cindy Wike). I did not have
this problem with Kai C++ v3.2 (non-RPM distribution), v3.3, or v3.4 on my RH
5.0 (not RH 5.2) system.