I would say that there are two big advantages for Applications…
First, when you write an application you can group the graphs however makes the most sense to you. That MSSQL check_mk script for example seems to have a few groups for its stats. Connections, Buffer_Manager, Buffer_Node, General_Statistics, etc. If you wanted to have all of the Connection stats as different series in one graph and all of the Buffer_Manager stats in another, an app can do that. With services, you just have one graph with one series per service check.
The second is of course is that application are as you mentioned auto discovering. You can use the Service Templates feature to do some dynamic groupings for Services. But if you write an app the whole community can use it and any LibreNMS user who drops in that check_mk plug-in will be able to get graphs.