Pixel value ordering (PVO) is a widely used framework for reversible data hiding (RDH). As the demand for higher embedding capacity continues to grow, achieving a proper balance between capacity and image quality has become increasingly important. In this paper, we propose a novel PVO-based multi-pixel embedding RDH scheme for grayscale images, which improves capacity by embedding multiple bits of data within multiple pixels in each block. A PVO recovery strategy is designed to guarantee reversibility while minimizing image distortion when multiple bits are embedded per block. Moreover, an improved flexible spatial location strategy is introduced, which defines pixel positions within a block using twelve modes. By selecting the optimal mode for each block, the number of expandable prediction errors is increased, further enhancing embedding capacity. In addition, the artificial lemming algorithm (ALA) is employed to optimize embedding parameters, enabling a better balance between capacity and visual quality for a given payload. Experimental results demonstrate that the proposed method achieves significantly improved embedding capacity while maintaining high image quality, offering a well-balanced performance compared to similar PVO-based schemes.