The unstructured mesh (UM)–based Monte Carlo (MC) method has gained significant attention for its adaptability to complex geometries and seamless compatibility with multiphysics coupling simulations, offering distinct advantages over conventional constructive solid geometry–based approaches. However, when tackling large-scale problems with numerous UM elements, memory bottlenecks may arise, limiting the practical application of MC simulations. In this study, a delta-tracking–based domain decomposition scheme tailored for UM-based MC simulations is proposed. This approach has been implemented in the MC code NECP-MCX, utilizing a bounding interval hierarchy framework to improve the scalability of UM-based MC simulations. Additionally, a run-time–evaluated processor allocation strategy was developed to automatically mitigate the deterioration of computational efficiency caused by imbalanced workloads. The developed code was validated using the Light Water Reactor Pool Reactor Benchmark and the Virtual Environment for Reactor Applications Core Physics Benchmark (Problems 1, 3, and 4). Furthermore, the effectiveness of the load-balancing optimization strategy was assessed using the Kobayashi Benchmark (Problem 3), known for its deep-penetration characteristics. The results were in good agreement with reference data and demonstrated reproducibility across simulations. In problems exceeding the per-processor memory capacity, the proposed domain decomposition scheme outperformed the sole domain replication scheme in efficiency. Moreover, this efficiency advantage was sustained across various scenarios, including deep-penetration problems, owing to the automatic load-balancing optimization strategy.