If tabular data is already in memory, what’s the point of having a cache at all? Memory is memory, right? Both are in main memory and access speed is the same, right?
Good question! Yes, access speed is the same. However, there are other benefits to a cache store.
For example, even though the data is already in memory, queries against the tabular model can still be slow… very slow even… usually in cases where the execution of the query is bound to the single threaded formula engine. To be sure, this is not a tabular specific problem… formula engine bound queries can be (and are commonly) found in both technologies and the issue (usually) speaks more to the deign of the model and/or the way the query is written (be that DAX for tabular or MDX in multidimensional). That said, performance problems related to FE-bound queries can be terribly difficult to resolve as it usually requires a redesign of the data model and rewrite of the query or measure(s) involved.
Bill points out the limitations of this solution, but within those limitations this looks like it could be a huge time-saver for end users.