It is good to see folks trying different approaches. Occasionally it leads to a new solution, but it always leads to learning.
A gentle balancing force should be adequate if the unbalancing forces are also gentle. It fails when the unbalancing forces exceed the weak balancing.
For well balanced packs it should work okay.
When a more serious balance problem develops it will quickly overwhelm the balancer and the weak cells, and if not detected the weak cells will be destroyed. You might say this is okay since the cells are bad, but this is not the whole story. For example, in a 4P pack, one of the three cells can go bad, resulting in the destruction of the three good parallel cells. A better system will catch this condition and allow the user to replace the bad cell and not damage the others.
Combining weak balancing with a cell level high voltage cutoff for the charger would solve this issue and provide real protection.
The question is, when used with a pack level charge cutoff, does weak balancing provide more pack life or protection? Since it does not detect the problem, it would not provide any real protection. It might provide a little extra life by gently "steering" toward balance. The same thing is accomplished by periodically charging with an RC balance charger, "resetting" the pack balance, with the opportunity there to "learn" about the balance of the pack. I suppose the monitoring can occur here, more or less, if the pack balance is checked periodically with the cell meters. So the user becomes part of the BMS.
If one of the weak balancer circuits fail and get stuck on there is no indication, and it will kill the pack. Unless you happen to monitor it frequently enough to see the problem.
In some sense, using a weak balancer is like what lithium manganese chemistry does, providing a weak restoring force toward balance. This does seem to work, but not as well as a true balancer (I heard Makita recently added balancing to their LiMn tool system and bad pack incidence went way down).
In summary, this technique should produce some improvement over no balancing, and some risk of circuit failure without detection, and not nearly as much improvement as full balancing.
Very interesting work!