Ryan Gregory has posted another interesting discussion about junk DNA [Genome size, code bloat, and proof-by-analogy.]. You should read the entire article but I want to comment briefly on two important points.
Computer simulations rarely tell us anything about real biology in spite of the fact that computer scientists think otherwise. The main problems with most simulations is that they assume a pre-defined goal and they usually don't model random genetic drift and other non-adaptation events.
The second point is ...
Finally, it is essential to note that "non-coding elements make future evolution possible" explanations, though invoked by an alarming number of genome biologists, contradict basic evolutionary principles.