Ok, I'll just steal that one from Jack's University of Batteries, but I do think it holds true. Swelling is telling and if your cells swell you are damaging them. Hence there is no need to bind them together tight, unless you want to destroy them by overcharging or, oh, let's say by Top Balancing, which really is just a form of doing that.
Another very, very interesting thing is the excellent paper from Sweden Jack is also kind enough to present to us. What's apparent is that to get the 2000-3000 cycle cell life (down to 80% capacity) you need to continously charge to 100% and discharge to about 10%. I think you'll get the same results by Top Balancing as well if you boil your cells at 100% SOC for extended periods of time. And every time you charge to 100% you're harming the cell, so it's really better to Bottom Balance and undercharge a little, say by charging to 3.5 volts per cell. It's the best way to take of your cells really. You only charge them to a voltage and stop immediately.
The test went on to 9000 cycles with LiPO cells in cyclic simulated EV use before the 80% capacity limit, if you didn't charge them over 50%. It didn't matter how fast you discharged. They lasted better if you went from 23 degrees Celsius to 35 degrees. I know it's LiPO cells, but they should be similar enough for all this to apply to LiFePO4 cells as well. Nothing whatsoever in there that would support anything the BMS people are claiming.
I'd go as far as claiming that if you only have 80% capacity left after only 2000 cycles you've probably been doing something wrong. If you are using a BMS then it's your BMS that is harming your cells.
Another thing to take home here is that discharge rate doesn't matter, which means I'm definitely going to remove any artificial current limits in my motorcycle right away. The SE40AHA cells can put out 400 A, that's 10 C, and I'm going to take it all. My AXE7245 controller could go up to 450 A. I will add some cooling though, I have a 8" fan and a 3" marine blower that I'll be experimenting with, but more on that later.