Making more "computerish" moves can't win a game if playing even a single bad move afterwards. Such bad move can be exploited by players like Anand even if you continue to play the best defending "computer" moves. So the rule assumes a shape and pattern - don't play a bad move and try to play always "the first line"!..
Btw, if you go through most of the Carlsen's games armed with an engine, you'll see that he's not playing "the first line" all the time. On the contrary, he often plays the second or even the third "computerish" line (e.g. the most "human" one). And you'll find that he plays a bad move very, very rarely. That's the point!.. And his opponents often say in interviews that he puts some constant pressure - that's why he wins so much games.

I wish there could be some sort of function where you could estimate an Elo rating from one or a few games, telling the computer also what the time limit was. This way you could compare your own games from totally different time periodss or rating system from a standard base. You could also see if a group of players was apparently performing at a level above the ordinary. Sure it has a lot of flaws, but I think it would be interesting/useful. I have no idea why Chessbase, who are a commercial organization, doesn't include this as a function that something like Houdini doesn't have. It couldn't be that hard, they could even put a "beta" stage in it. It could also give you your estimated ICC rating, estimated fics rating etc., and you could clearly see for yourself what the difference is. 

As for comparing computer evalutions and graphing them vs increasing Elo... I would not take the median because a computer could take some moves it doesn't understand yet and give them a slight minus evaluation which would lower the median, even if the player later made a move that shifted the computer evaluation to +4.0... and the player only gets the credit of shifting the median up one a half a place? It would be greatly biased against players that came up with one swift or amazing attack, where the sensational move would only move the median up half a place!!! I also am unsure if this graph would be useful or interesting anyway, there is no reason to think it would be linear and if it were quasi-linear, it wouldn't mean anything anyway.

Well, I believe that chess is an art. It's often difficult to tell the quality of a game without being at least an expert. The better the player, the more likely that player is to be able to judge the beauty of a game. 

I think I could determine whether a fairly large (perhaps 20 games) random selection of games were played by a 2200 rated player or a 2600 rated player. If I had access to a computer, I might be able to tell the difference between a 2500 and a 2700 rated player with a large selection. 

Beauty in chess can be created by almost anyone. Granted, the most beautiful games are almost always played by +2500 rated players, but I'm sure that most of the contributors here have created their own minor masterpieces.

If you see a draw between two players ranked about equally, can you tell if it was 2500 vs 2500, 2300 vs 2300 or 2000 vs 2000? Similarly for a won game between two players with a ranking difference of 200, does their absolute rating tell accurately? Perhaps one game is not enough, but a 2 or 4 game minimatch? Just something I've been thinking about.