True simulation requires tons of data, not just time. For example, different soils and rocks erode differently, and to simulate it one would actually have to have a more detailed model of the terrain than possible. It would be easier if the whole planet was made of sand or one type of dirt, but it's not (lots of erosion demos work just with that). Means we'd have to get detailed data for rock and soil composition, then also get good enough models for their erosion and then get lots of time to run erosion algorithms to produce good detail.
But we do not need to simulate it to get believable results. What we have got is a map of rock & soil types around the world, and we can get samples of how the erosion looks like on them, and we can (hopefully) devise an algorithm that procedurally generates the desired patterns.
It would be different if we did purely procedural planets, where the large scale erosion pattern isn't given by the elevation data but must be generated. But even there, I lean more towards pattern-matching than to the simulation, because of its efficiency and better compatibility with artistic input. For example, imagine somebody draws a fantasy planet, outlining continent shapes and mountain ranges. You can run erosion simulation to break the mountain range into peaks and to produce the river network, but it requires many passes and the erosion model usually doesn't contain nearly enough information to produce interesting results globally.
Also, artists may want to place major rivers there that suit the world setting, and want the rest of the world be refined automatically. A reverse-erosion algorithm would take these input data, compute river basins and iterate from the river upwards, carving the terrain in the process. Basically saying - if the river goes here, the surrounding terrain must be like this, to be realistic. The reverse process enforces patterns (with procedural alteration) to produce something believable. Doing so it needs a lot less resources and computation passes than the simulation.
