UniRestorer++ introduces a scalable universal image restoration framework built on multi-granularity degradation modeling that progressively refines degradation groups and attaches group-specific experts, coupling model capacity growth with specialization. Instead of simply increasing backbone depth or width — which leads to performance saturation — or relying solely on degradation-aware specialization vulnerable to estimation errors on out-of-distribution and real-world inputs, the method employs degradation-estimation-based routing to select a root-to-leaf path through a hierarchy of degradation groups and granularity-estimation-based routing to choose an appropriate level along that path, balancing specialization and robustness. Spatially dense degradation representation maps further enable region-aware routing for spatially non-uniform corruptions. The authors construct CDD-50, a benchmark spanning 50 degradation categories from single to penta-degradation, to evaluate scaling behavior. Experiments reveal approximately log-linear PSNR trends across the evaluated model-size range, with stronger scaling gains on out-of-distribution and real-world data than vanilla mixture-of-experts scaling. UniRestorer++ achieves state-of-the-art performance on standard benchmarks and demonstrates zero-shot generalization and few-shot transfer capabilities on real-world images.