>I completely agree here. But think you can't reasonably expect to get 50,000 records in a very short time. The problems discussed in my article are relevant if you fetch few records, out of a total of very many. You can still benefit from partial optimization, if you fetch 50,000 records out of a total of, say, 1 million.

In the example I posted to Jay the most restrictive case is actually about 200,000 records. I have another example where I need to issue a group by and from about 1m records and I need to join to a very small table.

The group by returns about 50,000 to 60,000 records.

Combining all of this into one SQL out performs the separated partial optimization.