The instructions are absolutely clear. Run to completion on the first node, and then (if you wish) in parallel subsequently. If you run on two nodes concurrently to begin with, you will have to de-configure both nodes and run again. I have seen this situation (in the classroom) a few times.
Oracle Certified Master DBA
On an 8 node cluster you could do:
Run root.sh on node 1 wait for it to finish
Run root.sh on nodes 2, 3, 4 ,5 ,6, 7 wait for all 5 to finish
Run root.sh on node 8
Given that on most cases root.sh only takes a second or two to run it is probably just as easy to just rn it serially on all the nodes.