Slush bases his pool speed off of the speed of getwork requests being made (as far as I can tell)
I'm not so sure.
The consensus on IRC was that pool performance is measured by submitted solutions, and I've never seen slush contradict this.
Do you have some evidence / numbers / URLs?