Thanks Rich for your answer. Just some questions:
The bug that caused this to happen was fixed, but unfortunately cannot
fix the bad nsState that already exists.  The problem is that the CSN
generator attribute (nsState) in the cn=replica entry for the suffx is
not cleaned up properly when you re-init replication.  In general, you
can't do this, because you could generate CSNs that you have generated

I think the solution here is to first unconfigure replication,

Must I unconfigure all the replicas, or just for the database giving problems?
shutdown the servers, then dump the database(s) to LDIF,

I suppose I must export only the data (with db2lif), not the replica information (with db2ldif.pl -r) because the server has shut down and this last script requires the server to be running.
then remove the
nsState attribute.

When exporting the data, has any entry that attribute? Or from where remove that attribute?
You will have to do this on every server.  Then,
start up, reconfigure replication, reload the data,

What do you want to say with "reload the data" and after "re-init the other replicas"? Import the exported data in one server, and then re-init the rest of the servers?

and re-init all of
the other replicas.  Make sure all of your servers are in time sync
before you begin.

Yes of course.
I know this is a pain but I don't know any other way to get rid of the
bad nsState.
> Regards and thanks in advance.

Regards and thanks in advance (again ;)).