So lets look at the graph again:
I’ve drawn three red boxes. See how when you saturate the links you get latency spikes and packet loss?
The trick is to have a mechanism that detects it and stops using a link which is having a spike of latency for a moment then use it again when latency is better. You can do this with latency Difference Cuttoff:
Which you can turn on in the profile:
Click save and apply on that and then share the result.
Also speedtest.net traffic is optimised by the ISPs (it is their key KPI after all). You want to look at how much usable bandwidth there is between your device and the Fusionhub using WAN Analysis as the real measure of usable bandwidth :
WAN Analysis - WAN to WAN Speed Test