I haven't actually done this myself, but I've run into a situation where it cropped up. We eventually "deep sixed" the offending server and rebuilt a fresh copy.
Let's assume that you have 20 servers participating in merge replication to support a high transaction volume web-site. You periodically take some of them offline for maintenance. When five servers are down, you make a schema change. Four of them come back online later in the day, but one has more serious problems so it stays offline for another four days while you get parts.
When that "prodigal child" comes back online, it still won't know about the schema change. In a high transaction volume situation, you almost certainly don't have five days worth of replication saved up. Bingo, Houston, we have a problem here!