The key thing to remember when adding a cell in parallel (or battery in a bank) is the impact at the top and bottom end of the ranges when charging and discharging. So, if your pack is dedicated to balancing, you can control all that. The issue arises when you throw one in a currently cycled pack "in production", where you no longer have the ability to control it to achieve balance, and are just blindly hoping it all works out. You have a good shot in that scenario it will all work out, but are taking a little bit of risk when it hits the top or bottom end of the pack charge cycle.
Assuming you can dedicate the pack/bank to balancing and have plenty of time, what I do for my bank is cycle with a very small load. I chose to stay near the top of the range, because my load was so tiny compared to the size of the bank. I also charged each to a similar V before putting in parallel to speed things up. I periodically let it rest by turning the load off. The only reason I used any load was to just get the electron blood flowing a little bit. Eventually, I removed the load and only went back and forth between rest and a little charging until they reached an optimal initial SoC at rest (13.6V with LiFePO4, which will eventually drop to 13.3xV at rest).
The risk if they are off and then you charge to full SoC is that you can overcharge the higher outlier. And, vice-a-versa, if one is lower voltage than the rest of the pack, and you discharge the pack to a low SoC, you risk the outlier entering damaging territory below it's ideal bottom.
In between, charging and discharging, or just keeping idle, will lead them to become balanced, presuming all other things being equal. But, if they have very different internal resistances, different self-discharge rates, etc,... then extreme cycling can actually make it worse. In other words, balancing will be more optimal with lower ranges of cycling. This is why we say that when we top balance, our goal is to keep it in the upper part of the range for awhile, because it needs time up there to really level out. Just going to 100% SoC then quickly back down will not achieve top balancing.
To be honest, the biggest risk with parallel is a bad cell or battery. They can drain the rest of the cells bringing them down to an unhealthy SoC. This is why it is more important to identify the capacity, internal resistance and self-discharge rates of cells before adding them to a pack. Self-discharge rate is the most important because if there is a problem -- typically an internal short -- this is where you'll identify it.
Unfortunately, a BMS cannot identify a bad cell in parallel. Only in series. This is one of the caveats of parallel.
