Well... climate is important, and Wilbur doesn't care about that.
Also, because of the way the flow accumulation algorithm works, low-gradient areas tend to make laser-straight stream courses. A real river would meander, but there's none of the dynamic feedback effects in the Wilbur model.
Also, the straight flow accumulation model often creates a lot of long parallel stream segments. In reality a main stream segment would tend to pirate together a lot of those. Deposition effects can produce yazoo tributaries, which parallel the main stream for awhile, but not usually great rafts of them.
For the most part, Wilbur will give you a good sense of where streams will lie given the terrain, but you have to give the terrain a bit of a grain of salt. Erosion is only of limited help: incise flow uses the same flow accum algo and can sometimes compound the areas, especially on flat areas. Doing a few long and short -distance precipiton runs and some small-amplitude magnitude noise can make it look better. I'm trying to perfect a method of selecting flat areas and digging a meandering seed trench, but it's slow going. Definitely, use the Wilbur rivers, but be ready to play with them.
One thing I do is to look at the river tex between noising and incise flow. Use the noise brush where the rivers look too straight, add another bout of absolute magnitude noise and fill basins. Look at the river tex again. If it looks good, apply incise flow. Otherwise rinse and repeat. This can be really slow, but probably a good deal faster than a dynamic river model with feedbacks.