Linear regression under log-concave and Gaussian scale mixture errors: comparative study