Is the Potential Energy just a bookkeeping device? It is said that if the space is homogeneous then momentum is conserved. But I've been thinking in the following situation:
Consider a parallel plates capacitor. In between the plates there is a uniform electric field so that the space is homogeneous. There is no point in between the plates that differs from any other. However an electric charged particle would feel an electric force and its momentum would change.
I know that the potential energy (PE) is not uniform, but at same time I understand PE as just a bookkeeping device (In the sense of section 4-1 of Feynman's lectures of physics, vol 1). Is this true or it has an underlying reality (in the sense PE existence cannot be avoided). 
Edit: Another way to put it. Suppose we never used the concept of energy in physics. We only use forces, momentum or anything we can measure directly. How could we say the space in between the plates is not homogeneous so that momentum does not conserve?
 A: I am answering the question formulated after the "edit" in a newer version of the text because that one seems well-defined.
Indeed, a situation with a uniform field $\vec E$ may be said to be "uniform" or translationally invariant in space. Noether's theorem says that this "uniformity" (spatial translational invariance) implies the existence of a conserved quantity, usually named momentum.
How does it agree with the fact that in this uniform field, the momentum of the (charged) particle is not conserved? Well, it's a different momentum.
The conserved momentum resulting from Noether's theorem will not be $m\vec v$ as the normal one but rather $m\vec v-qt \vec E$ where $q$ is the charge and $t$ is time. The time derivative of this quantity is zero which will simply say that $\vec a$ is whatever constant is dictated by the field.
I discussed the electric field but it is a more general result.
In terms of potentials, one may obtain the electric field $\vec E$ from a space-independent potential $\Phi$ (the freedom to make this true may perhaps be described as the potential's being a bookkeeping device – but what we really need is the gauge invariance) as long as the vector potential $\vec A$ is time-dependent (linear in time). The term $qt\vec E$ is nothing else than $q\vec A$ (up to a sign I am not sure about). Quite generally, in electromagnetism, one distinguishes the kinetic and canonical momentum that differ by the $q\vec A$ term. This is also important for the quantum mechanical description of electromagnetic and especially magnetic fields.
