Consider the pipeline

db.example.aggregate([{$match: {x: 4}}, {$skip: 1000}, {$limit: 1}])

In this pipeline, the $skip stage forces a split in the pipeline (since it would be incorrect for each shard to apply the skip). Then we're left with this execution plan:

mongos> db.products.explain().aggregate([{$match: {x: 4}}, {$skip: 1000}, {$limit: 1}])
{
	"mergeType" : "mongos",
	"splitPipeline" : {
		"shardsPart" : [
			{
				"$match" : {
					"x" : {
						"$eq" : 4
					}
				}
			}
		],
		"mergerPart" : [
			{
				"$skip" : NumberLong(1000)
			},
			{
				"$limit" : NumberLong(1)
			}
		]
	},

So as you can see, each shard has no idea that it actually only needs to produce at most 1001 documents to satisfy the query. In this case and in others where the $limit comes after stages which do not change the number of documents in the pipeline - we should duplicate a $limit stage on the shards part of the pipeline.