In a Web-based Augmented Reality (AR) application, to achieve an immersive experience we require precise object detection, realistic model rendering, and smooth occlusion in real-time. To achieve these objectives on a device requires heavy computation capabilities unavailable on most mobile devices, this can be solved by using cloud computing but it introduces network latency issues. In this work, we propose a new network protocol named DLDAR (Dynamic Load Distribution in Web-based AR) that facilitates and standardizes methods for dynamic division of compute between client and server based on device and network condition to min-max latency and quality.