Now that we wait for stream positions whenever we do a HTTP replication hit, we need to be less brutal in the case where we do timeout (as we have bugs around this).
