We revisit the effectiveness of land use densification as a strategy to promote travel sustainability by investigating the nonlinearity and mediation in the effects of residential densities on home-work distance, car ownership, and driving distance. Using a 2017 travel survey dataset in Beijing, we adopt a quadratic generalized multilevel structural equation model to estimate nonlinear and mediating effects simultaneously. Results suggest that the total effects of densities on home-work distance, car ownership, and driving distance are significantly curvilinear. Over-densification may result in marginal or even countervailing travel consequences. The appropriate level of residential density should lie below a threshold at around 18,000 persons/km². Moreover, the densification policy in low-density neighborhoods may have a trivial direct effect but a significant indirect effect on driving reduction, through the mediation of other travel decisions. These findings demonstrate the importance of incorporating nonlinearity with mediation to evaluate land use policies, particularly in high-density cities.