If a full-merge join is performed in serial mode and there is no need to sort the tables, the cost of a merge join on T1 and T2 is the sum of the cost of the scans of both tables, as long as all join values are unique:
The cost saving of a merge join over a nested-loop join is:
For a nested-loop join, access to the inner table of the join starts at the root page of the index for each row from the outer table that qualifies.
For a full-merge join, the upper levels of the index are used for the first access, to position the scan:
On the leaf page of the index, for nonclustered indexes and clustered indexes on data-only-locked tables
On the data page, if there is a clustered index on an allpages-locked table
The higher levels of the index do not need to be read for each matching outer row.
