You're simply squeezing a bit more memory bandwidth out of some nodes compared to others.
If you set them all to 1333 then they would all run at 1333, but the timings may be different.
In what way would the timings be different? Would it be better to have all 130 nodes set to "auto" and let the CPU manage things?
The end goal, for this particular group, is to have all nodes as close as possible when it comes to performance. They are running on all of these nodes using a mathematics algorithm and there's a contest which determines code performance based on run time. I've been asked for optimal BIOS settings so that all nodes are as close to the same setting as possible, even if RAM is slightly different.