This work presents a multi-task reservoir computing framework based on virtual node partitioning (VNP) within a single optical-feedback semiconductor laser. By partitioning the virtual nodes of a solitary delay loop into distinct subspaces—each dedicated to an independent task liker chaotic time-series prediction—the system enables efficient multiplexing of hardware resources. Modeled using the Lang–Kobayashi equations, VNP is employed to assign readout weights, thereby maintaining a unified reservoir architecture while simultaneously executing three Santa Fe benchmark prediction tasks, each achieving a normalized mean square error (NMSE) below 0.0023. Following parameter optimization, a single laser concurrently processes three different chaotic input streams—the Mackey–Glass sequence, the Lorenz sequence, and a laser-generated chaotic sequence—while maintaining high predictive accuracy. The minimum NMSE reaches 0.000043, and each sub-reservoir exhibits a memory capacity exceeding 25. Compared with single-task operation, the multi-task configuration introduces a marginal increase in error, demonstrating strong parallel-processing capability. Unlike prior approaches requiring multiple lasers or feedback loops, the proposed VNP scheme enables efficient multitasking without additional hardware. By eliminating the need for multiple active components, this approach not only reduces system complexity but also holds promise for a compact and scalable architecture in photonic reservoir computing, thereby laying potential foundation for future embedded applications.