The significant storage demands of 3D Gaussian Splatting (3DGS) constrain its practicality in real-world applications. Many compression methods using vector quantization suffer from low codebook utilization and prolonged training time, particularly for large-scale scenes. To address this, we propose HilComp, an efficient and training-free compression framework designed for post-training optimization of 3DGS models. HilComp employs a contribution-based Gaussian pruning strategy to eliminate redundant Gaussian primitives. Next, we propose a balanced clustering method based on the Hilbert curve, which leverages the spatial coherence of Gaussian parameters to compress them into compact codebooks. With quantization-aware fine-tuning and entropy coding, our method achieves up to 58× compression, 2.8× FPS improvement, and 65.6% energy reduction, while maintaining rendering quality comparable to existing methods. Extensive experiments demonstrate that our method achieves the best trade-off between compression efficiency and rendering quality compared to state-of-the-art compression approaches, while significantly reducing the compression time.