This quantity should be altered for each solvent. One way to get RSOLV is to try several values for RSOLV, and see which one gives the best solvation energies when compared with experimental results.
Dr Klamt suggests: For COSMO-RS, use a solvent radius which is independent of the solvent, e.g. R_solv = R_hydrogen = 1.3 Angstrom. Do not use a large solvent radius for large solvents, because the solvent is not probing the solute globally but locally. It is the local curvature which is relevant here, and a general value of 1.3 Angstrom is normally reasonable.