For static and low-speed applications, it doesn't matter. If you need full torque at speed, it's nice to have a datasheet that provides a graph like this:
image from: nanotec.com; note semilog scale
Roughly speaking, the effect of increasing the supply voltage from 24V to 48V would roughly double the RPM which is the "corner" on that graph.
The main reason is winding inductance. Magnetization is controlled by current, and rate-of-change of current is limited by max voltage. At the high end of the speed range, the lower supply voltage cannot increase the current fast enough.
For example, let's say you have a 200step/rev motor, 4A per winding, 2mH per winding, and 24V supply.
Max dV/dt = V/L = 24/.002 = 12000 A/s.
A very crude approximation of max rate needed is a triangular current profile (in reality the profile will be sinusoidal for effective high speed motion, but let's neglect it here for simplicity). This triangular profile has the current swinging from +4A to -4A and back to +4A, at each step = 8A swing per half step = 16A/step.
Thus the roughly highest rate at which a triangular profile is possible, is (12000 A/s)/(16A/s) = 750 steps/s = 3.75rev/s = 225RPM. This a crudely reckoned limit of where torque would start to fall off.
In real-world applications, at higher speeds it's also good to have excess torque capacity beyond what's "used" by the load, to let the system accelerate/decelerate and cope with resonances. Driver details would play a role there.