With archive and validation complete, this run is on track to finish in 25 minutes at a total time of 6.2 hours. The minimal time on the maxi has been 5.2 hours IIRC, so not bad. There's still a good amount of drift beyond the window, which started about 100k blocks later than the previous run, and this run will complete about 15 minutes sooner than that one. So the closer coupling actually reduces cost, and the only way that is possible (because the coupling has a higher cost generally) is that unnecessary paging is being reduced. The smaller the window, the less paging there will be, but the higher the coupling cost. So that will end up being the tuning point based on hardware.