How it works
The smallest feature parameter roughly corresponds to the smallest preserved geometric detail in world units after remeshing of the input data. Features smaller than this are likely to disappear during remeshing, although may still be approximated by textures. Smaller values will result in more detailed and accurate output assets while increasing output file sizes and computation times.
By default this parameter is set to 0.125: in this case the smallest preserved feature would be roughly the size of a cube with each side 0.125 units. This is however always in relation to the scale of the objects in the scene. Using the same smallest feature parameter for different sized content will yield different results.
The images below demonstrate the relative size of the Smallest Feature in comparison to a car model at original scale vs. scaled x 3.
Therefore the best strategy for obtaining the desired model quality is finding the perfect balance between Model Scale and Smallest Feature. Often this means scaling the model up for computation.
Won't my model then be huge in comparison with the rest of the scene?
Yes, but you can scale it down, but attaching it to an empty parent and scaling that down.
Below you'll see a few screenshots of the resulting model using a combination of smallest feature and object scale.