Ok, a few points..

Your changes make the *initial* XID random. The initial XID is set when
an RPC connection is set up. The rest of the XIDs is generated by
decrementing the initial XID.

If that avoids duplication, it's fine with me, since the additional
slow code is only used when setting up the connection; it's a one-time
overhead that you won't really see in a benchmark such as lat_rpc
or spray(8).

However, if you mean to randomize *every* XID, this algorithm is
way too expensive.

- Frank

-- 
Frank van der Linden                                            fvdl@netbsd.org
===============================================================================
NetBSD. Free, Unix-like OS. > 45 different platforms.    http://www.netbsd.org/