We study mutual-aid games in which individuals choose to contribute to an informal mutual insurance pool. Individual coverage is determined by the aggregate level of contributions and a sharing rule. We analyze theoretically and experimentally the (ex ante) efficiency of equal and contribution-based coverage. The equal coverage mechanism leads to a unique no-insurance equilibrium while contribution-based coverage develops multiple equilibria and improves efficiency. Experimentally, the latter treatment reduces the amount of transfers from high contributors to low contributors and generates a "dual interior equilibrium". That dual equilibrium is consistent with the coexistence of different prior norms which correspond to notable equilibria derived in the theory. This results in asymmetric outcomes with a majority of high contributors less than fully reimbursing the global losses and a signi cant minority of low contributors less than fully defecting. Such behavioral heterogeneity may be attributed to risk attitudes (risk tolerance vs risk aversion) which is natural in a risky context.