This is a good idea in theory, but in practise it would require obscene levels of processing power to achieve it. Each Peplink at either end of a VPN would need to proactively monitor and test and react to network conditions in real time, then dynamically build and teardown routing logic to initiate and deactivate WAN smoothing on groups of WAN links.
Of course it is possible, but to achieve it the Peplink devices would likely need more processing power, more RAM, be bigger and more expensive and we’d soon get to the point where they would be unattractive commercially.
Instead, I would suggest you consider looking at the latest SF feature which is outbound policy for Speedfusion. This allows you to set up multiple tunnels between two end points and configure each of those tunnels differently. So you could have a tunnel with WAN smoothing enabled, and another one with lowest latency set. Then only send your most important traffic via the WAN smoothing tunnel (like VoIP).
That way the bandwidth overhead of WAN smoothing is only on specific traffic types and not all of your traffic.