Coronary artery disease (CAD) remains one of the leading causes of mortality worldwide. Phonocardiogram (PCG) signals offer a non-invasive, affordable, and accessible means for early detection of CAD. However, the diverse acoustic manifestations of the disease across different auscultation sites make accurate diagnosis using a single-channel stethoscope challenging. Moreover, the scarcity of large annotated datasets further limits the development of robust diagnostic models. This work presents a multichannel CAD detection framework using transfer learning that leverages both early/late fusion from multiple auscultation sites. A lightweight pretrained deep learning model is designed to address data scarcity and enable computationally efficient deployment. We explore early and late fusion strategies to extract the channel-wise collective information in detecting CAD. The proposed system achieves a 9.46% improvement in accuracy over its single-channel counterpart, highlighting its potential for practical and scalable CAD screening. Clinically, it provides an affordable, accessible, and efficient tool for CAD detection, especially in low-resource settings.