Well, once I looked up what "empirical" means...I don't know if there has been such a study...probably depends on how you define it. But, I don't think there have been great earthquakes rupturing subduction zones that have been followed by another in same location before stress has had a chance to build back up. I suppose there may not be enough of a historical record for many subduction zone to know well the long-term history...and I don't know how many subductions zones have been studied with reasonably conclusive methods like drowned trees and tree rings (like Cascadia had been). I suppose one problem is that slip on earthquakes studied with modern methods is so variable...one patch may slip 20 m but the majority of the fault surface just a couple (Sumatra M9.3 for example). So, an earthquake could still start in some part of the earlier rupture before stress/strain has built up over the whole fault. But, I think I've heard that a rupture will not propagate too far into a part of a fault that is not near failure. So, I would not expect another M9+ to be likely in the same part of the subduction zone that failed in 2004-2005 for at least several decades. There was some mention at the SCEC meeting of it taking 10 earthquakes to release the stress in the rocks (or, was it along faults? Also, I don't know whether this was total stress of differential stress (and don't remember the difference between deviatoric and differential stress)...30 bar stress drops with 300 bars in the rocks...but along weak faults this may not be true ?). And, I doubt it works this way...that you can fire off multiple quakes on same fault patch without reloading.

A structural geologist would look at fault failure as related to stress perpendicular to fault holding it together along with friction, vs. shear stress ...differential (deviatoric?) stress parallel to the fault pushing it to slip. Fluid pressure makes the effective normal stress less and a fault weaker. In complicated tectonic areas such as California, perhaps stress loading is uneven and fluid pressures vary. There are new ideas about tectonic loading varying over 1000-5000 year time scales by a factor of 2 or so. But, for the Carrizo segment of San Andreas, it seems the modern loading of the San Andreas is the same as the paleo-seismic rate...about 34 or 35 mm/yr. If the paleo-seismic rate shows that the fault can break every 100 years and it has been 150 years, then should it have a 10% chance of failing in the next 30 years? Or should that be doubled, or tripled? Do you not think that the Cholame segment should have a higher probability than the Golden gate and north part?

By the way, 2 years ago I went on a SCEC field trip to the Wrightwoood Weldon/Fumal/(Scharer?) trench sites and the recurrence is likely solid but the slips and slip rates are shaky: base on complicated model that I find shaky.
Chris