I tend to work QVDs that match my end data model so that 1 QVD matches 1 table. If you're planning to load all the countries every time you load your dashboards, then a single QVD makes sense.
Subsetting makes more sense if you're sharing QVD across many dashboards, and each dashboard typically needs a subset. In this scenario, you'd rather not load the entire set and then filter down, but rather load the subset.
