I really haven’t encountered the trick that your friend is referring to, so I quite honestly cannot comment on it at all.
Being a computer programmer by trade, I find myself thinking like one … which basically means breaking down any problem into its component parts and figuring out some way to program a digital computer to do the right thing … whatever “the right thing” might happen to be. Therefore, when you tell me “there are two types of color in the eye when the light hits it at an angle,” my response is exactly as before: I want to turn it into an if-statement. 
“The final image that you see,” from the point-of-view of the computer, is, more or less, “the output of a computer program.” And by “a computer program,” in this case I really don’t mean “Blender, itself,” but rather, the output of the instructions that you gave to Blender in the form of various node-networks. (It could be more complicated yet than that … if you’re compositing, then “the final image” is the product of combining two or more intermediate results, each one of which was produced in some similar way, ad infinitum.)
Nevertheless… the way that I try to think of it (and, mind you, who-the-hell says that what I am doing is right, or that my crazy ways should be imitated by any other fool?) is: to break down the problem into its individual components (e.g. “(1) the color when the light isn’t hitting it, and (2) the color when it is”), and then to devise a rule for the computer to blend the components together, based on some set of parameters (e.g. angle of some plane vs. angle of camera) that I figure that I can actually get my hands on. I also want the final result to be adjustable, as many times as I may care to adjust it, without re-rendering.
For me, it is very definitely a matter of successive refinement. I never try to get everything to come together “in one grand-poohbah render.” Quite the opposite. There might be dozens of digital pieces all scattered about on the floor of my hard-drive waiting for some final composite to bring them all together. If your friend made his observation and I decided that it really would make a human-noticeable difference in the final result, I would consider how I could incorporate that finding into the final-output while re-rendering as little (if anything) as possible.
I’d try to say: “okay, this is great, and I am totally satisfied with it now except for my friend’s suggestion. Therefore, how can I incorporate my friend’s suggestion into this workflow while re-rendering little, or better yet, nothing at all? Furthermore, if I decide to discard my friend’s suggestion, how can I get right back to where I am now without having to re-do anything?” I’d like to get a “before” picture and an “after” picture so that I can easily and painlessly decide whether to keep it or to dump it.