A setting that looks fine on a photo may make a screenshot’s text hard to read. Testing a mixed sample seems more useful than chasing one impressive reduction.
How do you compare the results? Would you keep different settings for screenshots, product photos and illustrations?