Such clustering may prove difficult given building layouts and worries about theft. But the study underscores that often the best research comes from unscheduled interactions between scientists as opposed to scheduled research discussions and even the simple choices of placement of offices for faculty and students should take this into consideration.
Let me throw out another question. Is it possible to get these chance encounters in cyberspace? Can we have some sort of virtual complexity coffeehouse? Or does the Internet, despite its great properties of improving communication and distributing information, actually stifle innovation by preventing those random connections we need for research?