There's a group 3, but it's likely just as small or smaller than group 1. Where your data is easily normalized (and easier to work with in that form), but the cost of a getting an rdbms to support your write load is an order of magnitude more expensive than a persistent 'nosql' datastore that you do batch dumps into an rdbms.
The catch is that very few people end up in group 3 and still have the cost of running/administering the rdbms over the 'nosql' one actually matter.
Comments
There's a group 3, but it's likely just as small or smaller than group 1. Where your data is easily normalized (and easier to work with in that form), but the cost of a getting an rdbms to support your write load is an order of magnitude more expensive than a persistent 'nosql' datastore that you do batch dumps into an rdbms.
The catch is that very few people end up in group 3 and still have the cost of running/administering the rdbms over the 'nosql' one actually matter.