We have load generators for YCSB (just raw key-value ops in a firehose) and TPC-H (very complicated read-only queries) running right now, and we're about to start running TPC-C queries (moderately complex queries in large volume) as well. You can follow along on our progress here: https://github.com/cockroachdb/loadgen
In the context of your dichotomy, we want to bridge that gap. We want the linear scalability of your second group along with the full feature-set of the first group.
We will be publishing our performance numbers, but we haven't so far because the product has improved rapidly, and our numbers have been quickly obsoleted, but rest assured, we will be publishing a series of blog posts very soon. Anecdotally, our beta customers are not finding that they need very many more CockroachDB nodes than their existing database solutions, even with something as high-performant (but inconsistent) as Cassandra.