I've measured it at about 10 GH/s per 1C for SP30s, or around half what you indicate. I've measured this by taking a machine out of our racks (on around 14C air, IIRC) and sticking it into the coldest spot of our hot aisle (about 26C at the time, IIRC), and seeing about 120 GH/s drop in the hashrate. My memory is a little fuzzy on the exact numbers we got from that experiment, but it's accurate to +/- 20%. I've also done other measurements by looking at diurnal variation in performance on SP30s with similar results.

SP10s are much more sensitive to temperature. They also vary in performance by about 10 GH/s per 1C, despite having less than 1/3 the total hashrate.

I've heard that SP20s are more sensitive to temperature than SP30s, but I don't have any here to test.

Preliminary observations with SP31s (as well as predictions based on what I know of semiconductor physics) suggest that they're more temperature sensitive than the SP30s, but I haven't done any systematic measurements yet.