This could easily be the end user's Internet connection. It would be their WiFi connection.
You don't mention the type of VPN you are using.
If you are not using AutoVPN, could you buy a Z4 for one of the users? This will collect a lot of stats about the connection BETWEEN you, and also let you see the quality of the end users Internet circuit.
Below is an example of the kind of statistics that can be reported on when doing this.
Of interest is MOS - which is a measure of voice quality. You can also define performance classes when you do this.