> The performance problem of the queries where total result set is
> 2 million rows is not sorting, but simply retrieving all those
> rows. I wish that the "first"/"limit" type queries had never
> been invented and designers were forced to think about the selectivity
> of queries rather than just "what the heck, select two million and
> throw out one million nine hundred ninety-nine thousand seven hundred
> and fifty rows."

i agree, but if i could know when a query will return 2 000 000 rows before to launch it it's will be great ! unfortunatly no way ... without the order by everything i perfect but because of the #{^[\| order by it's a hell :(