Metaheuristic optimization algorithms are indispensable tools for solving complex engineering design problems, yet most existing methods suffer from premature convergence and an inadequate exploration–exploitation balance when applied to high-dimensional search spaces. The curse of dimensionality causes exponential growth of the feasible region, rendering algorithms that perform well at 30–50 dimensions increasingly ineffective at 100 dimensions and beyond. This paper proposes the Multi-Strategy Population-based Optimizer (MSPO), a novel metaheuristic that addresses these challenges through three synergistic search strategies coordinated by an adaptive selection controller. Strategy 1 Lévy-Enhanced Global Exploration (LEGE) employs Lévy flight-driven position updates with adaptive, dimensionality-scaled step sizes to enable long-range exploration of multimodal landscapes. Strategy 2 Opposition-Based Directional Exploitation (OBDE) combines opposition-based learning with a multi-reference directional exploitation vector that targets a convex combination of the global best, local neighborhood best, and elite centroid, thereby avoiding single-attractor stagnation. Strategy 3 Stochastic Dimensional Crossover (SDC) performs dimension-wise recombination with adaptive Bernoulli masks driven by per-dimension population variance, preserving diversity in unconverged dimensions while accelerating convergence in settled ones. A dynamic strategy selection controller activates strategies based on real-time population diversity, fitness improvement rate, and individual stagnation indicators. Critical dimensionality-scaling mechanisms automatically adjust strategy parameters for problems ranging from 10 to 1000 dimensions. MSPO is comprehensively evaluated on the CEC 2017 benchmark suite (30D, 50D, 100D), the CEC 2022 suite (10D, 20D), extended high-dimensional tests (500D, 1000D), and eight constrained real-world engineering design problems against 12 state-of-the-art algorithms. MSPO achieves the top Friedman rank across all benchmark suites, with the performance advantage widening significantly at higher dimensions attaining 2–5 orders of magnitude better convergence accuracy than the best competitor at 500D and 1000D. On all eight engineering design problems, MSPO finds optimal or best-known feasible solutions with the lowest variance across independent runs.