Effective anomaly detection in business process event logs is critical for maintaining operational integrity and ensuring process compliance. If left unchecked, anomalies from system malfunctions or unexpected interventions can cause significant financial losses and undermine subsequent process analysis. However, existing methods for multi-perspective business process anomaly detection often struggle to capture the rich semantics of event logs. Moreover, current attempts to leverage Large Language Models (LLMs) primarily focus on control flow, overlooking other important perspectives. In this paper, we introduce MALL, a Mamba-based Autoencoder enhanced by an LLM for multi-perspective business process anomaly detection. MALL leverages an LLM to create semantically rich embeddings, while using a Mamba-based architecture as its sequential backbone. Notably, its key innovations lie in a bidirectional Mamba-based encoder that captures full contextual information, and a specialized decoder where the control flow perspective guides the modeling of other attributes for more nuanced detection. Our comprehensive experiments on both synthetic and real-world datasets demonstrate that MALL significantly outperforms existing methods.