Here is using window of radius 3, sprinkling them apart, and copying the whole window over from source texture, blending the overlap areas.
The optimization runs a lot faster, but I don't think there is an increase in results. I think I am missing something here. I ran some tests with many iterations, and the results just kept looking like noise. I also have box/grid marks that are not blending correctly when trying this method of copying the whole window over.
It's late. I'm calling it a night.