I am not sure if this is a regression or if the original bug SERVER-2249 was not fixed completely.

If the field already exists then the duplicates are correctly filtered.
—
db.test.insert({_id:"test_add_to_set",arr:[]})
db.test.update({_id:"test_add_to_set"}, {$addToSet:{arr:{$each:[1,1]}}})
db.test.find()

{ "_id" : "test_add_to_set", "arr" : [ 1 ] }

—

If, however, the field is new then all of the values are added.
—
db.test.insert({_id:"test_add_to_set"})
db.test.update({_id:"test_add_to_set"}, {$addToSet:{arr:{$each:[1,1]}}})
db.test.find()

{ "_id" : "test_add_to_set", "arr" : [ 1, 1 ] }

—

I was able to reproduce this in version 1.8.1.

Cheers,
Leif