The basic problem is that balancing is generally round robin (the router will assign a connection to the next WAN in the list, then the next, looping through the list constantly), so the more WAN connections you have, the more likely it will be unevenly loaded, especially when they're all different speeds. Of course this also depends on if single-thread throughput will just hit max, then it should just work, but especially on mobile networks that's not a given.
In pfSense I did have weighting, that basically changes the allocation so the faster connection is say 5x more likely to get used than the slower one, but you still can't account for the fact that's not based on a given client on the LAN, its just how it allocates traffic over the WHOLE LAN when a new connection is requested. So when trying to test the full speed, you wont necessarily get the allocation you expect as you add more WANs.
Plus different speedtests use a different number of connections, so you wont necessarily get the ideal number of connections to each WAN to max it out. It becomes somewhat pointless given real-world downloads will be either a single connection, or maybe 4 for downloaders like Steam. (I really should check how many it REALLY uses)
So it seems to work really well with 2 WANs, pretty good with 3, somewhere along the line 4 just caused an in-balance where I couldn't get all four to max out on a test. Of course 4 WANs is pretty extreme and it was just while I was testing the best combination to stick with, when my second DSL line contract was coming to an end so I was trying a few different mobile plans to see if they were sufficiently stable to replace it.
My original plan was once I got FTTP to have that and two mobile plans, one which stays at home as a backup, one which I could unplug and take with me when I go to meetups, I could pick which one based on coverage of the network where I would be going. But as the Pandemic never actually ended, I don't go anywhere so it became moot.