For an intuitive example of where this is more efficient, consider plane changes. The delta-V for a plane change is 2 * sin(angular_change/2) * speed

The higher an orbit is, the slower you're going at apogee. If you're doing a large plane change, going even slower will save you a *ton* of delta-V, so you can transfer up to a much higher orbit, plane change at apogee, and then transfer down to your desired final orbit.

That's by far the most useful place for bielliptic transfers, not just going from one orbit to another on the same plane.