Alternatively, maybe it has nothing to do with physics, the picture is computer rendered, and the artist found the first of these two alternatives a whole lot easier:
1. model a train that is straight, and then use some tool to bend the entire model to fit a curve.
2. define a curve, then align models of train engine and cars so that they follow the curve.
If the train is elastic, you'd recover that energy.
Going into a curve, kinetic energy from the speed of the train is turned into spring energy by bending the train. Leaving the curve, the spring energy becomes kinetic energy. The flexed train speeds up due to the spring force as the train unflexes.
I don't think this is correct as a general statement. Imagine if the walls were made of canvas fabric with appropriate amounts of slack, just as an example of a material that can easily bend.
You would need a cleverly designed frame, but
I'm sure there is something that would work. Perhaps many smaller segments joined together.
Canvas can bend more easily than rubber or steel, but deforming it still requires some energy. And while canvas would work for the walls, it wouldn’t work for the bogey, and anything rigid enough to work as a bogey would take lots of energy to bend.
To some extent you can say a regular multi-bogey train is undergoing deformation by changing the angular configuration between cars via the linkages. This could technically be extended to have the linkages have every few inches along the bogey instead of between 20-foot segments. We just need a way to transfer energy reliably between deformations and change in acceleration.
1. model a train that is straight, and then use some tool to bend the entire model to fit a curve.
2. define a curve, then align models of train engine and cars so that they follow the curve.