Trajectories are critical for Location-based services, yet they frequently contain long-range, irregular gaps caused by sensor limitations and environmental constraints. While existing recovery methods rely on self-context and neighbor-context modeling, they often fail to capture fine-grained motion dynamics and perform poorly under sparse data. To address these limitations, we propose TMV, a Traversability-enhanced Long-range Trajectory Recovery Framework with Motion-variation Modeling that jointly models individual motion patterns and collective movement conventions. Our framework incorporates a motion variation-aware trajectory encoding module that quantifies speed and direction changes across gap boundaries, guiding an attention mechanism to enhance high-frequency transitions and mitigate over-smoothing. Additionally, we develop a traversability-enhanced grid encoding module that rasterizes trajectories into spatial grids and employs a transition-aware masked autoencoder with local neighborhood attention to learn robust inter-region movement patterns from sparse data. A density-guided fusion strategy dynamically integrates these embeddings, prioritizing grid-based collective behavior in low-density regions where self-context becomes unreliable. Extensive experiments on three real-world datasets demonstrate that TMV achieves state-of-the-art performance, particularly in complex urban environments with limited trajectories, outperforming existing baselines by 18.67% in Hausdorff distance.