Simulating water flow for the whole world would be a demanding task, and it wouldn't necessarily lead to the same state of the world as we know it now, because of the limited resolution of elevation and precipitation data, and because of complete lack of historical data for such simulation. Since the elevations we've got are computed from the lengths of the reflected rays from satellites, and the rays happily reflect from water surfaces, we don't have the river beds there. So, the end result would be quite different.
Our plan is to use a vector database for the main rivers, but a kind of simulation might be employed for small streams for which there are no readily available vector data, but that can be inferred from the terrain itself. Running the simulation on a smaller scale, basically just by identifying the drainage basins of the smallest rivers from the vector dataset and simulating the flow on them separately might just work.
It's also something that we would like to do for a planet designer, where you start with a planet by outlining the continents, mountain ranges and arterial rivers, and fractal-driven algorithms would refine it into details. Unlike the main rivers, course of which largely depends on initial chances and weather during the ages when they formed, small rivers better follow the simple downhill flow rules. So while you could largely affect and determine the course of a large river, small rivers will flow as they must, or else you would also carve the mountains anew.
