In the scale of thousands. For example in ECS, where you have hundreds of entities, you may wish to store the entity data in a more compact way. Or in another case, if you wish to transfer data across the network faster.
In any way, it will be slower than the native table indexing operation. This is due to the speed in which the buffer reading methods operate. At the moment, the module uses a proxy object to make the indexing system work. If buffer metatables were to be implemented, then the overhead of the proxy object would be removed, thus allowing faster indexing.
That aside, at the moment, per 1000 objects, the indexing operation takes around 100 microseconds. (0.0001 seconds) Versus 10 (0.00001) from native table indexing.