In distributed multimedia applications, content is often delivered to users in a degraded form due to network-induced lossy compression. Real-time and interactive use cases like cloud gaming, which render content on the fly, require low latency and are hosted at resource-constrained edge servers. We present a new insight: when rendered content is delivered over a network with lossy compression, high-quality rendering can be ineffective in improving user-perceived quality, leading to a poor return on computing resources. Leveraging this observation, we built Stimpack, a novel system that adaptively optimizes game rendering quality by balancing server-side rendering costs against user-perceived quality. The system uses a mechanism that quantifies the efficiency of resource usage to maximize overall system utility in multi-user scenarios. Our open-sourced implementation and extensive evaluations show that Stimpack achieves up to 24% higher service quality and serves twice as many users with the same resources compared to baselines. A user study further validates that Stimpack provides a measurably better user experience.