Here's a picture of what I'm talking about, using an example of potential problem meshes. (where there are these kind of flat areas, large 'elongated' polygons)

And the difference between correctly importing the normals from MoI in the left, and letting the app itself compute the vertex normals based on regular methods in the right.