Don't know how seismologists refer to "gaps", but I would use the term more loosely. John V.'s post here was interesting, and I have run across talks also that it does not work, and especially that characteristic EQs don't work...perhaps the first time was a talk at UCSB by Ralph Archuleta sometime around 1990 +/- couple years that the Parkfield prediction was flawed. John's post goes along with the work by Weldon-Fumal et al that the San Andreas fault at Wrightwood goes through periods where it slips much faster than others. But, I went to the site they used for much of their data in a field trip, and it seems that it is very solid that there are periods with more large earthquakes, but that the modeled amount of slip on those quakes is very questionable. Weldon et al's GSA Today paper more or less said that if there has recently been a quake on that part of the San Andreas, it may be more likely, not less likely, that there would be another one. That would probably require what John V. said here...that the stress drop in quakes in only partial. But, I saw a talk by Mark Zoback a long time ago (or read a paper, or something) that at Loma Prieto 1989 the stress drop was enough to allow left-lateral aftershocks on faults parallel to the San Andreas there...meaning that the shear stress along the fault for right-lateral was entirely released (hopefully I got the terms correct...I need to look up deviatoric stress and differential stress because I don't remember the difference...but I am still operating my home office out of boxes and all is dusty and I have gotten t where I really avoid looking in the boxes.
Chris