[plorkyeran]

If the in-memory DB has the same feature set as the real DB, using the in-memory DB for tests should be a single flip you switch, while writing code to generate fake result sets for every tests requires writing all of that code. 90% of the benefits for 1% of the cost.

[jeltz]

But they do not have the same feature sets. And PostgreSQL trivially be tuned to work almost like an in-memory database.