This turns on bilinear filtering of the composited screen buffer used as the colour source for the depth-of-field effect (and its downscaled version).
This simply means that we can get away with around half as many samples for the same quality of bokeh effect.  The samples we do take express the desired radius more accurately too, so there aren't such glaring delineations in blurredness.

To celebrate being about twice as fast, I've also upped the maximum fatness of the bokeh effect for situations of extreme defocus.  So typical cases are around twice as fast, and the extreme bokeh situation is around the previous speed but with fatter highlights.

Been running like this for a while myself...