A major factor is the lack of context. When the people conversing are colocated, they share a context and their conversation respects their context. When the people conversing are talking over data pipes, they do not share context and their conversation subsequently does not respect the context.

For example, when I am a passenger in a car, I typically pause the conversation momentarily if the driver is trying to change lanes or exit the freeway. If I'm walking down the street with a friend, I might pause and stare at a clown instead of dumping a constant stream of audio into my friend's ear.