Given a subtree of entries spread over several servers and linked using referrals. When performing a subtree search, that follows referrals,
some errors that occur are ignored.

The problem occurs when following continuation references that in turn
generate further continuation references. Currently, the references are
processed depth-first. However, as the continuation references are processed
any errors that occur are lost.

The effect is the no error occurs but the number of entries returned is
less than expected.