A shared repository is usually used for distributed scanning/indexing (i.e. if scanning happens at once capture station and indexing needs to happen in another). In this scenario, it is imperative that the batch is shared in a common repository. Based on my personal experience, it is a safer option than non-shared because in the non-shared repository, there is a chance that the mdb file gets corrupted more often than not.
FileNet is correct. When you install the instance of your CE, you would already be having a DB server. You need not create a separate database for a shared repository, You can use the same database create a DSN (in the machine where you want to configure Capture) which points to this DB. While you are configuring the repository in Capture Manager, you can point it to this DSN. Yes, based on the volumes of your scan, the table will get populated with the repository specific data.
Hope this helps