Yeah... I really don't understand a lot of robotics research. They seem to be forever chasing these awkward "proof-of-concept" implementations of concepts that are completely uninteresting. This is a perfect example: obviously you could make some robots that could do this, but it's really unclear what you'd learn by doing so, and the result is useless.
I mean, if they actually wanted this behavior for some purpose, and this was a reasonable way to approach that practical purpose? Sure, do it. Of course. If there was some question whether this was possible (which obviously there wasn't, at least not serious), then sure, prove it's possible. If there was some question we'd be answering (about biology, maybe?) or some challenge that would be interesting to overcome, then, uh, maybe. But it seems much more likely that such questions could be answered easier by a simulation.
As it stands, it seems like we've got a HUGE surplus of research on, to pick a random example, how to have robots all go somewhere without bumping into each other - most of it saddled with arbitrary restrictions that are also completely uninteresting. Can the swarming robots co-ordinate without any predetermined communication protocol? Yes, obviously, and we can prove that in simulation. But that's not good enough for some reason, have to actually build some crappy robots to mong into each other while we re-solve a bunch of boring practical problems with batteries and light sensors.
If you just want to have a challenge, or get some practice building robots that have to deal with the real world, why not build them to at least attempt something interesting or useful, or at least entertaining?