I created a sharded cluster with 1 mongos, 1 config server and three shards each with a single node replica set.

I pre-load the data and the contended table in question has this index:

{D_W_ID:1,  D_ID:1, D_NEXT_O_ID:1, D_TAX:1}, {unique:true}

in addition to _id index. The collection is sharded with shard key

 {D_W_ID:1, D_ID:1} 

The following two contended operations start majority of transactions:

db.coll.findAndModify({D_W_ID:1, D_ID:1}, /* query */
     {$inc:{D_NEXT_O_ID:1}}, /* update */
     {sort:{D_NEXT_O_ID:-1}}, /* sort */
     {fields:{_id:0, D_NEXT_O_ID:1, D_TAX:1}}, /* projection */
     {new:false}
});
Ddb.coll.findAndModify({D_W_ID:1, D_ID:1}, /* query */
     {$inc:{D_SOMEOTHER_FIELD:1}} /* update */
});

With readConcern “majority” every once in a while one of these two findAndModify commands matches 0 documents, which is impossible.

There are no inserts, deletes or balancing happening to this collection, only lots of findAndModify and a very occasional find.