The code to reassemble multicast fragments has serious bugs in its design and
implementation.  First, it assumes that a remote process will not get reassigned
the same host:port combination after a crash, this is clearly incorrect.  Second
the implementation uses 32-bit unsigned integers, but then performs dumb loops
like this:

for (CORBA::ULong i = purge_first; i <= purge_last; ++i)

this is an infinite loop if purge_last is std::numeric_limits<CORBA::ULong>::max()

to add insult to injury I think the code handles its circular buffer incorrectly.

I have fixes for most of these bugs, and will submit a patch shortly.