On a second thought, using

 size_t          size = (size_t)(lrand48() % (LARGEST_BUFFER + 1));

instead of

 size_t          size = (size_t)(drand48() * (LARGEST_BUFFER + 1));

would have saved all the trouble as lrand48() is guaranteed to
return a result in the range of 0-2^31 in the lower bits and 
clean upper 32 bits.

This works on Linux (i386,PPC), too.
-- 
Volker Stolz * stolz@i2.informatik.rwth-aachen.de * PGP + S/MIME