It is, of course, quite easy to find the best completely-disjoint route. In particular, you can simply find the best route, remove all of the links in the best route from the network (or set their costs to be arbitrarily high), and then re-calculate the "best" route. The problem with this approach is that when all of the links in the best route are removed it may disconnect the origin from the destination (i.e., there may no longer be a path from the origin to the destination). In addition, even when the network does not become disconnected, the completely-disjoint path may be too dissimilar to be useful.
One is then naturally led to consider iteratively removing the links in the original best path. However, this can be quite expensive computationally because you actually must remove all combinations of links (e.g., 1 only, 1 and 2, 1 and 3, 1 and 4, ..., 2 only, 2 and 3, ...).
Instead, the following demonstration uses a method known as Lagrangian Relaxation to solve the problem.
Instructions:
Some interesting examples: