Of course you can formally make any scale transformation you wish, even "local" scale transformations. However, the equations of motion will acquire "fictitious forces" stemming from the non-constant transformation you have applied.
Let's take as an example a 3-body problem, in which we are looking for the motion of a light third body, and have the other two bodies travelling in a mutual circular orbit. Then scaling to use $\vec{s}_3 = \vec{x}_3/|r_{12}(t)|$ makes good sense and gets rid of a non-vital parameter; $r_{12}(t)|$ remains constant so no mysterious apparent forces arise.
But if bodies $1$ and $2$ are in a mutual elliptical orbit and you re-scale, using
$$
\vec{s}_3(t) \equiv\vec{x}_3(t)/|r_{12}(t)|$$
you will find that the equations of motion for the third body have become much more complicated. It would be much simpler to work in a scale where the semi-major axis is taken to be one; that is constant, and makes your work easier.