>> I thought that the Carrara image showed that the screen grab from MoI worked--despite the "theory"?

Well at that rendersize, you can use any kind of greyscale pattern and it will look right.
But if you're goal is to learn how create good displace/normal map, the method is wrong.

You can probably come to something close in this example if, when you grab the screen, you use a front light (actually you dont), and you should also hide the
wireframe. But if you do any closup or -mid closup, you'll see that the sphere shape will not look like sphere anymore, but more like some kind of spike.

Anyway you'll never be able to grab any subtile shape into a displacement map (or normal map) with that method, using 8bits value image from any 3d viewport lighting value.

As for the photoshop normal plug, it's a great helper to extract normal information based the grayscale value of an image (there are different method in the plugin), and can work quite well for high frequency details ... but if the actual grayscale is actually wrong, the normal will not look beter.

It's quite different that baking the normal from an highres geometry like x-normal can do.