Thanks for all of your responses.

This method has been helpful to us for purging log tables and history in our data warehouse. I realize that partitioned tables would have been a better way to handle history purges, but we were not able to implement partitioned tables for various reasons. (Maybe Phase II...)

As for why @cnt had to be DECIMAL: Say you had 11 rows to delete, and you set a TOP value of 5. Obviously, you would need 3 iterations of the DELETE to capture the last (1) record. However, when figuring out how many iterations you need, 11/5 = 2.2, which is a DECIMAL (which you use CEILING() on to get the next largest INT). AFAIK, in order to get a decimal result (2.2), the dividend itself must be of type DECIMAL.