The keys produced by the mapReduce emit function are always compared using the simple collation rather than the mapReduce operation's collation. For example, consider the following script:

db.c.drop();
db.c.insert({_id: 1, str: "foo"});
db.c.insert({_id: 2, str: "FOO"});
db.c.insert({_id: 3, str: "bar"});
db.c.insert({_id: 4, str: "BAR"});

db.c.mapReduce(
    function() { emit(this.str, 1); },
    function(key, values) { return Array.sum(values); },
    {out: {inline: 1}, collation: {locale: "en", strength: 2}}
);

This script produces output such as the following:

{
	"results" : [
		{
			"_id" : "BAR",
			"value" : 1
		},
		{
			"_id" : "FOO",
			"value" : 1
		},
		{
			"_id" : "bar",
			"value" : 1
		},
		{
			"_id" : "foo",
			"value" : 1
		}
	],
	"timeMillis" : 133,
	"counts" : {
		"input" : 4,
		"emit" : 4,
		"reduce" : 0,
		"output" : 4
	},
	"ok" : 1
}

Since the operation is using the English case-insensitive collation, one might expect the results array to contain 2 elements rather than 4. This is because "FOO" == "foo" and "BAR" == "bar" with respect to the collation.