Wind power forecasting (WPF), as a significant research topic within renewable energy, plays a crucial role in enhancing the security, stability, and economic operation of power grids. However, mid-term forecasting faces a persistent dilemma: achieving high predictive accuracy often comes at the cost of computational efficiency. Existing Transformer-based architectures struggle with this trade-off: traditional temporal attention mechanisms suffer from computational redundancy and weak inter-variable coupling, while recent transposed architectures, despite improving speed, inherently compromise the capture of local temporal dynamics and domain-specific periodic characteristics. To overcome these limitations, this paper proposes Fast-Powerformer. Built upon the Reformer backbone, the model reconstructs the feature extraction paradigm through three complementary strategies: (1) an Input Transposition Mechanism that optimizes multivariate coupling modeling while reducing sequence complexity; (2) a lightweight temporal embedding module that compensates for the intrinsic deficiency of transposed architectures in capturing local sequential features; and (3) a Frequency Enhanced Channel Attention Mechanism (FECAM) that exploits spectral information to characterize the physical periodic patterns of wind power. Experimental results on multiple real-world wind farm datasets demonstrate that Fast-Powerformer achieves the best overall performance among compared methods. The model successfully balances superior accuracy with reduced resource consumption, highlighting its significant practical potential for resource-constrained scenarios.