According to the notion of the "uncanny valley", after a certain point, more accurate renderings of humans become not better, but creepy, and that one has to approach complete accuracy to overcome that effect.
On the other hand, I recall a story about 19th century venture capitalists. It may be apocryphal, but here it is: Alexander Graham Bell was looking for money to fund further development of his invention, the telephone, and so he did a demonstration for people who might provide that money. Did he have Mr. Watson go to the other phone and speak? No, he had Watson go there and sing popular songs of the day. Why? Because the VCs already knew those songs, and would mentally fill in gaps and glitches with the words they already knew should be there.
So, when we see our friends in Second Life, which of the above two apply, if any? I'm inclined to say the latter, at least for those of us who've invested ourselves in SL (and posing the question in terms of friends presupposes we're of that sort). We want to immerse ourselves in this virtual world, and we're willing to overlook the inconsistencies. But what about the newcomer to virtual worlds? I wish I knew.