Ensemble Temporal Prediction Model-as-a-Service (ETP-MaaS) has become crucial in areas such as financial modeling, weather forecasting, and cloud monitoring, managing a dynamic set of base models and workers. Real-world systems face a two-fold challenge of dynamic collaboration and heterogeneity that current methods overlook. At the model level, data volatility dictates that optimal performance requires identifying and weighting constantly shifting subgroups of base models, not just individual ones. At the system level, these model groups must be efficiently mapped to a pool of heterogeneous and dynamically available workers. Existing solutions fail to co-optimally address these intertwined tasks, treating models as independent entities and employing simplistic allocation rules, resulting in poor accuracy and resource inefficiency. To this end, we introduce WIET, an efficient ETP-MaaS system that innovates in weight distribution and worker allocation to tackle these dynamics. We model evolving group behaviors among base models and propose a novel group temporal locality-enhanced multi-label classification method for highly adaptive weighting. Additionally, we develop an efficient, multi-dimensional worker allocation method powered by hybrid heuristic optimization, effectively reducing bottlenecks and resource waste. Extensive experiments have shown that WIET consistently outperforms state-of-the-art methods in terms of model accuracy, latency, and resource usage across various workloads and prediction tasks.