the main sense merge it is identical weeding , this normal desire for any merge
This normal desire the developer should designate uniqueness explicitly, setting uniqueness of indexes. That helps the server to select the optimal plan. And if indexes by the task nature are unique, but the developer hammered in a bolt to tell about it to the server the server selects not the optimal plan. For merge join leads to spooling in tempdb that strongly increases I/O. For other merges I do not remember, that uniqueness was critical.
it is passed...
Really important characteristic
Yes, unconditionally when it is a question about I/O. In the table all data successively, and in an index only the necessary fields. Even if the index is incorrectly sorted also speech about it , reading of an index and reading of all table is absolutely different amount of logical reads. And a difference that more than more than fat columns in the table. Therefore I always estimate the size of the table and the size of an index to understand, whether gives a scoring creation of a covering index.
In the given task the index on t2 most likely is used cannot be in search plan on it because of these or t2.field = ". Therefore it is not important, whether as it is sorted and sorted generally. There it is possible easily
CREATE INDEX IX_t2 ON t2 (PK) INCLUDE (field1, field2, field3)
and it changes nothing. Nevertheless, from the point of view of I/O all the same it occupies less places and less logical reads demands, than table reading t2 entirely.