If there are say three profiles in an IOR. A thread using this IOR find that the
first and second profile were unusable in the first invocation. It uses the
third profile to start communication. After a few invocations the third one
fails. The thread instead of trying out one and two again (possible that the
servers are alive now), returns a TRANSIENT exception. 

The funny part is, the next invocation could start communication to profile 1,
since we rewind before throwing an exception. 

This is a mistake.  A thread finding a dead profile should try other profiles at
the sametime before giving up instead of depending on old data!

RTCORBA solves this by going over the profile from first for every invocation.
We can do better for normal CORBA. FTCORBA is a different ball game all together.