There are many different ways in which learning can occur, with some appearing to offer unique advantages over others.
Hands on experience has the appeal of engaging the subject extensively, however the flipside of this is a greater cognitive load that can inhibit the ability to 'take a step back' from the task and gain a greater overview. For more on this see the work of Bandura, 1977.
Recent research by Hoover, Giambatistaand and Belkin (2012) compares direct experience and vicarious learning approaches to learning and concludes that an integrated approach is best, with vicarious learning preceding direct experience.
Of course there may be some tasks where one style is preferable over another; they suggest trainee surgeons would do well to watch and learn that way first before experimenting on real people. But equally, vicarious learning alone can have its limitations if the participant is not getting a full range of, for example, tactile experience.
A fusion of the two is likely to make sense in many instances, and the use of simulation can lead to a blending of the two types of learning. Think of a Matrix-like construct that offers an interactive quality that a participant can both make sense of and contribute to in a way that builds skill.
In such an environment, either technologically simulated or imagined, there would seem to be room for a continuum between observation and participation that could be adjusted, rather like a difficulty level on a computer game, as is suitable to a given user's current skill level. Perhaps some of the more difficult processes would be automated in the 'easy' version- an ideal ratio between immersion and cognitive demand thereby being reached.
In the mean time (whilst we are busy programming our constructs) a simple take-away from this is to both overview a task being completed by others and directly engage in it for better performance.
Read Hoover, Giambatista and Belkin's (2012) full study here.