Currently, the order members of a sharded cluster are shutdown in is:
1) Config Servers
2) Mongos
3) Shard members

See https://github.com/mongodb/mongo/blob/master/buildscripts/resmokelib/testing/fixtures/shardedcluster.py#L148-L171

If we change the shutdown order to be what's below, the we can shave ~1 minute off the shutdown time in some configurations.
1) Mongos
2) Shard members
3) Config Servers

We think this is because the shard members may try to contact the config servers after they've been shutdown which could lead to the shard members entering retry loop.