As described in the design document for sharded transactions, shards don't need to abort their local transactions on stale version errors if the router was stale, because the shard won't have to refresh its metadata in this case and won't need to take a collection exclusive lock on the collection that triggered the exception.