FWIW, if this is for collision detection, this is probably not necessary. Mesh/Mesh collision is not normally performed because it's impractical: When meshes consist of 100 polygons... that's 100x100 = 10000 polygon collisions you need to detect for just one collision. It's even worse when you have larger meshes.
There are shortcuts you can use to speed up the process... like the Separating Axis Theorum (wikipedia it)... but it only works with convex polygons so it's unsuitable for most meshes.
For most collision detection you don't need it to be "perfect", you just need to to be "good enough". One very easy technique is to form one or more spheres by having a central point and a radius for the meshes. Then to see if they "collide" you just check the distance between their center points and see if they are closer than the sum of their radii.
Low-res mesh collision is a common method, so I don't know that I would call it "foolish".
However, the two really aren't all that different in terms of accuracy. The mesh has the advantage that it can be rigged and articulated along with the higher-resolution mesh.