Suppose a request is made to a node that is lagging against the rest of the replicas with a recent afterClusterTime timestamp, this will cause the request to wait until that timestamp has been applied on the node.

Now let's assume that a DBA decides to remove that node from the ReplicaSet via a reconfig and doesn't shutdown the node afterwards.

Under this scenario the request made at the start could potentially never finish since the stale node is now outside of the ReplicaSet. This has potential availability implications if the request was an internal request made within the cluster since that request will potentially never finish.