A screenshot from the Beach Demo used by Google to demonstrate O3D's capabilities
Originally, O3D used a plug-in based architecture which allowed 3rd party developers to integrate custom functionality such as pre and post render effects, particle systems, and or physics engines for example. It is important to note that the plugin was written in C which communicated directly with the hardware, thus the speed of scene rendering was largely dependent on the graphics card of the computer rendering it. Now, much of this same functionality is built into WebGL.
The main advantage O3D has over alternative desktop or console based 3D rendering engines is that O3D may load, render, and transform models and their respective textures dynamically, using AJAX and/or COMET in real-time. Traditional compilation of source code, application resources, and object libraries is no longer necessary, since all of these aspects are loaded in real-time. These remote resources may be designed, developed, and maintained outside the core rendering or view application within a typical object oriented MVC application. The direct result of this, explicitly makes development of rich 3D application easier, as you do not need to recompile your O3D application per resource changes. This allows for a more robust and distributive approach when designing 3D applications.