You should leave the servo period at 1 ms (1000000 uSec)
For a step/dir system, There is no real advantage of going faster unless you have multi G accelerations
Also since there will be no base thread, a histogram with --nobase is more reflective of actual latency
Preempt-RT can be nearly as good latency wise as RTAI, it just depends on the hardware, BIOS setup, kernel version, kernel options
phase of the moon etc:
freeby.mesanet.com/h97-g3258-preemt-rt.png