The replication publish information about the whole topology in cn=monitor
When cn=monitor is searched, the replication therefore asks informations
about the replication state to all Replication Servers. This should always be
fast unless a server is hanged. In such case the replication waits for
5 seconds then issue an error message and goes on with the
information it has received at this time.

In some cases however it looks like that some server do not send back the
responses fast enough thus provoking errors in the logs.
I was not able to produce this problem but discovered some inefficiency in the
replication monitoring code that can lead to query several time the other servers
and therefore maybe producing the problem.

This change therefore improve the monitoring code so that we only wait once if
a server is very slow to answer.