vs
So the 1st is more preemption points, while the 2nd is making all kernel code preemptible, even outside of preemption points. And it states in addition to “slightly lower throughput” also “slight runtime overhead to kernel code.”
Generally, as far as I understood, “preemption” means to interrupt a process/thread to do other things in between. Usually, this can be done only at defined parts in kernel code, the preemption points. So the first options adds more of those points, so that processes/threads can be interrupted more quickly/at more stages, for other applications to run more quickly/smoothly under loader, while the 2nd option is much more consequent, allowing them to be interrupted anywhere.
This however should not add idle load, only changing behaviour under load 
. But can still be tested, of course.