Most modern in-memory database systems rely on multi-version concurrency control (MVCC) to support real-time data analysis without interfering with concurrent writes. While MVCC elegantly enables lightweight isolation of readers and writers, it also generates many short-lived versions and results in long version chains. Existing version storage designs employ traditional data layouts lacking version information. Consequently, as time elapses, the long-running query latency increases due to increased chain lengths. In this paper, we introduce a novel version-aware data layout for in-memory MVCC databases. This layout improves search performance for heterogeneous workloads. It differs from previous approaches by organizing data in a version-centric manner, which supports index-only searches and reduces costly version traversals. Specifically, the layout utilizes a version B+-tree to organize the latest tuple versions of a table, making the table index-organized. Additionally, it employs range partitioning and epoch partitioning techniques to group older tuple versions efficiently. This design allows for fast retrieval of the correct versions visible to transactions, minimizing redundant memory accesses. Compared to existing version storage schemes, the version-aware data layout considers workload access patterns and the opportunity to skip unnecessary version accesses. The experimental evaluation indicates that the version-aware data layout can outperform the state-of-the-art designs by 1.15× ∼ 2×.