Blockchain sharding technology is widely recognized as a key solution for enhancing the scalability of blockchain systems. However, traditional sharding systems often use lock-based cross-shard submission protocols to manage cross-shard smart contracts, which limits performance due to complex communication and consensus requirements. Some approaches attempt to shift cross-shard smart contract execution off-chain to bypass cumbersome on-chain consensus processes, but they still encounter significant communication overhead. In this paper, we propose a novel blockchain sharding system—EchoShard. This system efficiently executes cross-shard smart contracts through a collaboration of on-chain and off-chain processes. We design an off-chain execution node based on a trusted execution environment (TEE) that can effectively simulate and execute smart contracts of varying complexity. Additionally, we develop a new data structure specifically for managing data related to cross-shard smart contracts within the system, which further accelerates off-chain execution. Furthermore, we introduce a new cross-shard submission protocol to ensure that off-chain executed cross-shard contract transactions are submitted accurately and efficiently. To validate our system, we implemented a prototype of the EchoShard system and designed a series of smart contracts for comprehensive performance evaluation. The experimental results demonstrate that EchoShard achieves at least a 2.7-fold increase in throughput compared with traditional sharding blockchain systems. Compared with the latest proposed blockchain solutions, throughput is improved by 1.4 times, whereas the confirmation delay for cross-shard smart contract transactions is significantly reduced. These findings underscore the effectiveness of our system in enhancing the performance and efficiency of blockchain systems.