ADR is a tricky thing to get right. While on set, an actor focuses purely on their performance, but once they get into the ADR booth, a lot of their focus has to go into the timing, because they have to match an already finished performance. It locks out a lot of the potential emotion. On set, an actor has emotional freedom (well, depends on the director, but for the most part he does). An actor can add a pause between words, switch up some timing etc... and is fully able to express emotion that way. But in the ADR booth, it's sort of like saying, "I want to you act as though you have complete emotional freedom, like you did the first time, except that I need you to do it exactly like this, and match your original emotion perfectly, even if you aren't feeling that emotion any more." It makes it very difficult to express real emotion when you have such a large constraint. Some actors are very good at it though (like Brando and Crowe), and can make you believe in it easily.

Then on the technical side of things, there can be a few factors. It is important to try to match the type of mic used for the line on set, match (roughly) the distance from the mic to the talent, etc... especially if you are only replacing a single line or a word. Reason being that it is very difficult to eq a sound to match another one perfectly. There will always be a slight difference, and you can minimize the difference by trying to match the original setup as much as possible. Maybe only 1% of your audience can identify the difference, but more than that might feel it. If you are replacing a whole scene, the technical side (matching the mic) doesn't matter as much.

Last thing I'll mention is the room tone. When you record on set, you are picking up a slight (hopefully) amount of room tone in the mic. When you record ADR, you are in a "dead" booth, with no (again, hopefully) room tone. In order for the ADR to fit, you have to lay room tone (often called a "glue track") under the ADR to make it sound like the rest of the dialogue. This is assuming the sound mixer was even afforded the opportunity to record room tone (which often doesn't happen anymore, especially in faster moving productions). Even with the "glue track," it doesn't ever quite sound 100% right. It's something that you have to hide, or "burry" in the mix. This is something that almost nobody can pick up on unless you are trained to hear it, but subconsciously, your brain will tell you something isn't quite right.

Hope this helps a bit?

Interesting idea, nice one!

In my personal (but by no means extensive) experience, ADR by actors and directors that aren't extremely familiar with the process tends to produce performances that lack some character and emotion. The little imperfections that happen in the on-set delivery of lines also, i think, add realism. In ADR things can be over-produced to the point of being sterile.

On the other hand, some actors insist on ADR. I'm not sure how much material is out there on this topic, but i have heard the Marlon Brando and Russel Crowe deliberately deliver lines that will be unusable, so they can fine tune their performances in ADR.

Great question! and something that I'm personally very interested in too, and I have some random observations that you might find interesting. I can't really talk about ADR in Movies per-say but Games have struggled greatly with the recording and delivery of convincing dialogue since the creation of the video game, and a lot of the reasons behind this might translate into your question. The 'traditional' way of recording game dialogue is one actor in the studio at a time reading lines from the script - this is usually not very inspiring, creates a disjointed narrative and doesn't lead to a good performance. However games like the Uncharted series have recorded their dialogue straight from the Mo-cap stage. The actors get to interact and move when delivering the lines ergo a better characterised performance. There are also many varying degrees of recording practices between the 2 for-mentioned examples. So the conclusion is the closer the actor's voice is to the original performance the better that performance will be ... Even though neither of these examples are ADR I hope you can see some similarity with the problem of characterisation in relation to the context of the dialogue, I feel that for an actor being in the moment of the scene is key and the more intimately you can capture that the better it will convey to the audience.

BUT to be honest I think the final delivery is more down to the actor and how experienced/comfortable they are with the situation, ADR can improve a performance but it won't improve every performance. Interesting question though ...

This is a great subject to study, but it's definitely not going to be an easy one. The quality of ADR ultimately comes down to the individual actors. There is definitely an effect introduced by the recording engineer in terms of matching when it comes to tonal characteristics, and direction during ADR can be important too, but these both pale in comparison to what the actor brings (or doesn't) to the session itself.

Simply put, truly well done ADR (and editing of said material) becomes transparent to the audience. It's the instances where it is not optimal that it can be identified and affect the viewing experience. It's simply another potential problem point in a production; one that I can't fault directors like Altman for avoiding.