What you really want is the 'trombone effect'. Use a perspective camera, but 'fake' the ortho view by using a very small fieldOfView and set a long distance away. Then interpolate over time the fieldOfView to something normal like 60 and closer distance (or vice versa for reverse effect).

It goes something like this (my code is more complex so I'm just snipping the math out here.

Update:
  tmin = targetGameObject.renderer.bounds.min;
  tmax = targetGameObject.renderer.bounds.max;

  // we would like to keep tmin and tmax constant
  c = (tmax.x - tmin.x) / 2;

  if (Camera.main.transform.position.z > .0001 || Camera.main.transform.position.z < -.0001)
    Camera.main.fov = Mathf.Abs(4*180.0 / 3.14159 * Mathf.Atan(c/Camera.main.transform.position.z));
  if (Camera.main.fov < 0.01)
    Camera.main.fov = 0.01;
  if (Camera.main.fov > 180.0)
    Camera.main.fov = 180.0;

I found this code at the forums and it works like a charm: http://forum.unity3d.com/threads/32765-Smooth-transition-between-perspective-and-orthographic-modes