Monte Carlo rendering faces computational challenges in high-resolution visualization. While video super-resolution can be accelerated using low-resolution rendered sequences, existing methods exhibit information redundancy, suffer from error propagation in temporal fusion, and fail to exploit renderer-generated auxiliary features fully. To address these limitations, we propose the Auxiliary Features-Guided Cross-Frame Transformer for Rendered Sequences Super-Resolution (AFCT). Our solution integrates a Local Patch Pre-Alignment Module (LPAM), a Cross-Frame Local Attention Module (CLAM), and an Occlusion Perception Module (OPM). LPAM utilizes optical flow to pre-align sequence frames, minimizing information loss and providing subsequent attention blocks with highly correlated inputs. CLAM comprises an Inter-Frame Attention Block (IAB) for efficient local spatiotemporal modeling, which avoids redundant tokens across frames and enhances attention accuracy, and an Auxiliary-guided Attention Block (AAB) designed to fuse high-resolution auxiliary features for enhanced texture reconstruction. Furthermore, we employ the geometric information in depth to implement OPM, which dynamically adjusts inter-frame attention weights to mitigate artifacts. Experimental results demonstrate that our method outperforms six state-of-the-art methods in preserving spatio-temporal coherence, offering a quality-preserving acceleration technique for Monte Carlo rendering.