As I said before, I absolutely love the feedback node. You can cosmetically switch direction for better code layout, you don't need a dummy while loop for much cleaner code, and if you use the global initialization feature right, it can often eliminate the use of a "first call" primitive. You can place parallel FN in different cases eliminating tons of telephone wires (would even be better with this 11 year old idea, though. <hint, hint> :D).
Feedback nodes also don't have that abomination of resize-ability, leading to rigid code that never scales very well. Good riddance!
(Bob, not sure what a "feed-forward loop" is in the context of this discussion)
I'm pretty sure he's talking about the "switch direction" thing you mentioned above. I can use a basic feedback node, and there are some instances - like when your subVI lies within a loop and you need the benefit of a shift register but have no loop to hang them on - where there are no alternatives, but I always have to experiment a few times to set it up right. I imagine you can set it up without thinking, and frankly, it blows my mind. 😉