In recent years, sharding technology has partially addressed the scalability limitations of blockchain, resulting in the widespread adoption of sharding blockchains in domains such as finance and government-enterprise collaborations. However, existing sharding approaches presume node homogeneity, neglecting real-world variations in node network conditions. This practice may group nodes with poor inter-node connectivity into the same shard, causing degraded intra-shard transaction throughput and prolonged latency. While direct network-condition-based node assignment appears viable, it faces some challenges: ensuring accurate shard partitioning while maintaining system security through randomization. To address these challenges in heterogeneous network environments, we propose a reputation-based blockchain node reallocation (RNCR), comprising two core components: (1) Reputation-based node security assurance that using reputation to distinguish malicious nodes. (2) Network-condition-aware node reallocation that optimizes intra-shard performance by allocating well-connected nodes to the same shard. The method incorporates partial random assignment and malicious node removal to reinforce system security. Experimental results demonstrate that RNCR achieves approximately 40% higher intra-shard transaction throughput and 60% lower latency compared to baseline approaches, indicating significant optimization effectiveness.