Just the size of the drift does not provide any clue to understand if it is acceptable or not. Even the percentage variation with respect to the initial value is a meaningless quantity (although sometimes cited in the literature).

One has to remember that energy is always defined within an arbitrary constant and what is meaningful is to compare differences of energy with other differences of energy.

So, the drift you observe should be compared with the typical fluctuation of potential (or kinetic) energy. As thumb rule, you should have a drift of the total energy at least two orders of magnitude smaller than potential energy fluctuations in order to guarantee an acceptable integration of the equations of motion.

I am slightly worried that you have a drift at all. Runge-Kutta type methods do give a drift, but symplectic methods such as Verlet, or the corresponding 4th order methods do not. Rather they show oscillations in the energy. So I would at the very least suggest you consider looking at algorithms like Verlet (for a description check Wikipedia)