Actually it's almost possible. QV always sorts the distinct dimension values not the 'appearances' in each sublevel.
The trick is to make them different for sorting.
The easiest way is to use straight table and add calculated dimension =aggr(StoreID&Month, StoreID, Month)
Then you can sort by StoreID and then in that dimension by the KPI1 value. Of course you have to make that column almost invisible.
If the pivot is a strong requirement, you have to modify the data model to add the field that is look like Month, but differs for each StoreID-Month combination. The dual() function can be useful for that. Then just use it instead of Month.
Don't forget to make the pivot read-only to not frustrate the user if he accidentally makes the selection in that field.