We have excluded the use of API since the start.
The single solution is to replicate the data to have better performance. The software developed a lot the features to explore the historical evolution of the data and this meant a different architecture.
As you probably know, the difference between a database and a data warehouse is exactly the introduction of a third dimension (the time) that transforms everything in a cube. It means that the data increases a lot.
In addition, exactly for that, the data warehouse CAN replicate data to improve the performances. This replication and redundancy increases the database more and more.
We have spent two year to prepare the backend exactly to have a local replication of the data that would have been helpful also for other applications.
As you know, Wikimedia Ch developed also the Map Service together with Cassandra.
Kind regards