I tried the same comparisons with a real-world example, a data table with over 200,000 rows. In the first iteration the cursor method took 15 seconds, while I killed the 'array' method after 15 minutes after only a small fraction of the rows had been processed. The next time around I added a PRIMARY KEY constraint to the identity field and this did the trick. The time for the array method dropped to 8 seconds. Kind of a key omission, I'd say...