Is the insert performance achieved by reducing reliability? For instance, on Postgres you can speed up inserts significantly by disabling fsync, which means that a commit operation succeeds before the data is fully written out to disk. The article mentioned that this is now possible on SQL Server but does not mention whether the impressive numbers were achieved by enabling this feature.
Insert performance is only one aspect. Is query performance better? Not just simple queries that hit a single index or the primary key. Queries with multiple subqueries or joins will be interesting.
What about sharding? The cloud is all about scaling out and traditional RDBMSes are pretty complex to scale out while maintaining ACID compliance. Is this on the horizon for SQL Server?