Scott pointed out that it would be helpful to have a list of servers that didn't get the write and the OpTime that they are replicated to. That way, if w:something fails and all the other servers are an hour behind, the driver knows that there isn't much point in retrying. Conversely, if the other servers are almost up-to-date, it might be.

However, this is quite difficult because we don't keep a good map of local.slaves entries -> replica set members. local.slaves could have members in it that have since been removed from the set and we wouldn't want to list those in getLastError's response.

Also, we'll have to do something different for master/slave.