NoSQL vs RDBMS is really a proxy war for Denormalized vs Normalized storage.
You can take a system like Cassandra and treat data very much in a normalized way which would reduce performance. You can take a system like MySQL and completely denormalize your data which would increase performance.
Any test where one set of data is normalized and one isn't is not a fair test.
Also, denormalization can be a big deal. Unless you have some sophisticated code managing it for you, you're trading performance for data storage management complexity. Now you have to manage many instances of data X. But there is a benefit in that you avoid crazy joins.
I think both have concepts to learn from each other. For example, in order to use a NoSQL option effectively, you end up implementing your own concept of indexes, something very easily done for you in RDBMSes.
You can take a system like Cassandra and treat data very much in a normalized way which would reduce performance. You can take a system like MySQL and completely denormalize your data which would increase performance.
Any test where one set of data is normalized and one isn't is not a fair test.
Also, denormalization can be a big deal. Unless you have some sophisticated code managing it for you, you're trading performance for data storage management complexity. Now you have to manage many instances of data X. But there is a benefit in that you avoid crazy joins.
I think both have concepts to learn from each other. For example, in order to use a NoSQL option effectively, you end up implementing your own concept of indexes, something very easily done for you in RDBMSes.