This separation gives us a much-needed degree of flexibility regarding legislation. It allows one to make laws concerning just the concrete parts, or just the use of concrete parts in human-recieved communication, without needing to rule on the whole communication at once, which is very difficult and tends to overflow into other domains.
For instance, there are laws on the books regarding compulsory licensing of music, setting fair rates and methods of collection. Such laws have become a hindrance lately because of the way the expression concept conflates content with delivery. Thus, there is no one rate set for music licensing, nor even the possibly-feasible rate of licensing given an intended use (background to a radio news report, business Muzak playing, education and criticism, etc.). Instead, the rates are set for music given a certain distribution method, which is to say, the exact mechanics of how the message of the music is sent. Unfortunately, there are a wide variety of ways of distributing music, and subtle variations on each one. One recent example of this is how the law has handled the case of streaming music over the web, which went very poorly and upset lots of people.
Using this model, we can concentrate on just the important questions regarding the two ends of communication. What music was sent in a message, how was it used, and who received it? The exact manifestation of the "expression" is not what matters. It doesn't matter if the user is listening on the radio, or listening to a CD recording of some radio broadcast. What matters is how many people heard it, what the music was used for, and what music it was. Trying to enumerate all the delivery methods is doomed to fail, so don't try.
This drastically reduces the number of special cases in the law by eliminating the need to consider the large and rapidly increasing number of different media for delivery of a message, and would correctly handle an entire domain of concrete content, no matter what transmission methods or other uses are imagined in the future. So even though the new model is more complicated then the expression model, its ability to more accurately reflect the real world move us closer to all of the goals of simplicity, completeness, robustness, and usefulness.