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?
To be honest, I haven't looked in quite some time (a month or so) but I know that at one point, I was testing a client that ended up spamming a large number of getworks to his server for a few seconds, and I noticed the pool speed dramatically increase at the same time. It could have been purely coincidental, which is why I said, "as far as I know".