Spectral line shape models can successfully reproduce experimental Rayleigh-Brillouin spectra, but they need knowledge about the bulk viscosity ηb. Light scattering involves GHz frequencies, but since ηb is only documented at low frequencies, ηb is usually left as a free parameter, which is determined by a fit of the model to an experimental spectrum. The question is whether models work so well because of this freedom. Moreover, for light scattering in air, spectral models view "air"as an effective molecule. We critically evaluate the use of ηb as a fit parameter by comparing ηb obtained from fits of the Tenti S6 model to the result of Direct Simulation Monte Carlo (DSMC) for a mixture of Nitrogen and Oxygen. These simulations are used to compute light scattering spectra, which are then compared to experiments. The DSMC simulation parameters are cross-checked with a molecular dynamics simulation based on intermolecular potentials. At large values of the uniformity parameter y, y ≈ 4, where the Brillouin contribution to spectra is large, fitted ηb are 20% larger than the ones from DSMC, while the quality of the simulated spectra is comparable to that of the Tenti S6 line shape model. At smaller y, the difference between fitted and simulated ηb can be as large as 100%. We hypothesize the breakdown of the bulk viscosity concept to be the cause of this fallacy.