In this paper, we propose a new 3D sound rendering method for multiple sound sources with limited computational resources. The method is based on fuzzy clustering, which achieves dual benefits of two general methods based on amplitude-panning and hard clustering. In embedded systems where the number of reproducible sound sources is restricted, the general methods suffer from localization errors and/or serious quality degradation, whereas the proposed method settles the problems by executing clustering-process and amplitude-panning simultaneously. Computational cost evaluation based on DSP implementation and subjective listening test have been performed to demonstrate the applicability for embedded systems and the effectiveness of the proposed method.