Using Apache Arrow for data transfers between Spark and pandas minimizes overhead by enabling zero-copy reads, which significantly reduces the time and resources required for serialization and deserialization. This approach is highly effective in optimizing performance for large-scale data processing tasks.