Do some patrons want to know whether that name they are looking at is an author, illustrator, subject of a book? We are forgetting those that do have specific wants and needs. Any kind of a new coding system has to allow for methods of letting people know what they found, so they can decide if it is what they want.
Some differences make a difference in some cases; in other cases two items may be considered close enough, or near equivalents. Martha Yee's dissertation was a very good exploration of the subject. I believe that a lot of insight can be had into FRBR related methods by looking through the lens of some form of relative identity ("This is the same manifestation as that one, but they're different copies").
Maybe we need to be asking researchers what they need in order to identify the materials they need.
That's one possible methodology, and useful as a pilot study. However, the problem with is approach is that the SME may not have introspective access to this knowledge, or the answer may be "it depends". A better approach might be to test accuracy of judgement using a more experimental methodology (e.g. find minimal pairs of items that differ in just one feature, and ask "is this the same as that" or "when would you want to to distinguish these two things", etc.
Another approach might be to observe behavior in the wild, and match researchers' search activity with subsequent circulation events, then calculate whether the removal or addition of a rule would have changed the number of iterations, size of result sets, included checked out but not returned as a search result, etc.
With the right IRB incantations, and informed consent, treatment could be applied to live queries rather than retrospectively; this is necessary to evaluate catalog searches, which are almost always iterative in nature.
Of course, none of these issues ought to be relevant to bibframe...