Sometimes when the momentum residuals flatline above the default convergence criteria with oscillation (assuming an appropriate time step is used), it can be related to unsteadiness in the flow that's not properly captured running a steady simulation. It may be that using real fluid properties in this case is exacerbating this effect. It's always a good idea when encountering convergence difficulties to look at the solution in CFD-Post to try to understand why it's having troubles. In this case I would look at velocity vectors plotted on different planes to see if there are signs of flow separation. If so, try running the model in transient. If you get convergence running transient, it may indicate that the flow is unsteady. Using Transient Blade Row methods (TBR) you can generate an accurate solution that you can then compare with your steady, unconverged results to see what impact unsteadiness has on the performance parameters of interest.
If running supercritical CO2, I would strongly suggest using RGP tables rather than RK properties. The RGP table generater is now part of CFX-Pre at 2023R2 (under Tools->Generate RGP File). The advantage of using RGP tables is it will be more accurate, and the tables are designed to handle operation near the critical point with higher table data resolution near the critical point.