Your question is vague and I don't understand the underlying reason for doing this. Some PDB structures require additional changes to make them be in biological form, like dimers which are stored in the PDB only as a monomer. By limiting the structures to protein do you mean to take away anything which isn't a protein, including waters, DNA, ligand, and covalently attached structures? What if the structure contains multiple proteins? Multiple chains covalently bound with an S-S bond? How do you want to handle missing structures?
Which surface do you mean? The molecular surface, the solvent accessible surface, or something else? How do you want to handle pores and cavities? Consider in the extreme case the full shell of a hollow virus capsid, containing no DNA or RNA.
Does "cover" mean each sphere is completely outside of the surface or can it intersect the surface? Or that the center of the sphere must not be inside the surface of the protein? Can the spheres intersect each other?
It's impossible to specify both N and R and get an answer which covers an arbitrary surface. If N = 1 then R must be at least 1/2 the maximum width of the protein, if the sphere can be placed in the center. Otherwise there is no solution.
My guess is you're trying to do some sort of structural domain identification, but in that case having spheres is too limiting. I've seen commercial work (related to electrostatic surfaces) which use Gaussian ellipsoids. However, it's been too long since I've been in that field to know which tools are available.