Both of these configurations are equal. But now to find out which one is faster, gotta go to Cadence :(
From Wikipedia:
"A chain of XORs—a XOR b XOR c XOR d (and so on)—is true whenever an odd number of the inputs are true and is false whenever an even number of inputs are true."
Edit:
After math, the delay for the bottom "tree" circuit is 20.65 RC.
The delay for the top "waterfall" circuit is about 28.29 RC.
Therefore, the number of gates that the signal passes through matters a great deal and gates delay.
