If I get purist, using classes is like saying "drive it if it's a car". Using microdata is like saying "drive it if it has a Volkswagen logo". Presumably you could have something without microdata that is a "description", and vice versa.
It might also be more consistent if you only use classes for styling a segment, rather than a mixture of classes and microdata.
In your particular example, it probably doesn't matter. In a node that is higher up the tree these distinctions would be more consequential.