The tension between efficiency and equilibrium is a central feature of economic systems. In many contexts, social networks mediate this trade-off: an individual's network position determines equilibrium play, and social relations allow coordination on an efficient norm. We examine this trade-o in a network game with a unique Nash equilibrium, but such that agents can achieve a higher payoff by following a "collaborative norm". Subjects establish and maintain a collaborative norm in the circle, but the norm weakens with the introduction of one asymmetric node in the wheel. In complex and asymmetric networks of 15 and 21 nodes, the norm disappears and subjects' play converges to Nash on every node. We provide evidence that subjects base their decisions on their degree, rather than the overall network structure. Methodologically, the paper shows the capabilities of UbiquityLab: a novel platform to conduct interactive experiments online with a large number of participants.