I can only strongly support this idea, especially if we can define some kind of “cut-off quality” (combination of latency and packet loss) on a WAN connection in order to “deactivate” it in both Speedfusion tunnels as some outbound policy algorithms as described here: Cut-off latency in Outbound Policy priorities
Note however that the measurement of packet loss is defined by your interpretation: packet loss of the ping tests and the TCP retransmissions can be monitored, but plain UDP traffic will be dropped along the way without notice by the router.