Also: if your stated purpose is to create a medical video that must “model a physically-correct (healthy or diseased) human eye,” then all of these concerns of physical accuracy might well be relevant. But in all other cases, well, they aren’t.
Once you’ve decided upon all the shots in which a character’s eyes are going to be visible and subject to “extreme close-up (ECU)” scrutiny, then you can decide just how much detail is needed to carry the shot convincingly. Details of reflections across the cornea, the appearance of a believable iris, and (perhaps most likely) convincing “tiny eye movements,” will be most-important then, and, even then, “only up to a point.” There’s a point of diminishing returns, and it comes up fast.
You might well have several models of things like heads and faces. If the camera’s three inches away from the face as Our Heroine makes the brilliant realization that will Save The Planet, yeah, then you need an incredibly accurate eye. But if it’s a few feet away, though, using the same model (or perhaps, any model) will not “meaningly affect the pixels produced” to justify the CPU time consumed. Subtle nuances of her human performance will be what the audience actually sees . . .