I think the key issue here is that there is no necessary connection between fiction and reality to validate an exemplification. Sherlock Holmes is in fact a very potent case: he has many behaviours and characteristics not seen in most detectives, because he is a fictional character and not bound by reality. Thus it is fallacious to say that he exemplifies detectives because as a fictional character, he cannot be used to represent real things.
A more obvious instance of this is with animals. Suppose I describe to you a fictional horse. This horse does not exemplify real horses because by virtue of its fiction, I can attach to it properties like "runs at Mach 7" and "does calculus in his head," and this is perfectly valid within my context, while it certainly does not help to describe real horses.
Now, this may seem malicious on my part, and the natural response is "but so long as you stick to what's really possible, everything should be fine!" However, therein lies the problem: we have no reliable way of guaranteeing that our fiction totally reflects reality. No matter how hard we try, fictional characters may behave in unrealistic ways because our attempts to model their actions are ultimately speculative to some degree, unless we make them congruous to observed real world behaviour, in which case the character is no longer fictional.
Of course, this means that fictional characters exemplify fictional characters. Sherlock Holmes exemplifies fictional detectives, but here is where I'm not so certain: one might need to be careful by setting bounds depending on context. For example, Sherlock Holmes exemplifies fictional detectives in his fictional world may be the more accurate formulation. Batman is also a fictional detective, but I'd be reluctant to say that they both exemplify the same sort of fictional detective.