Under transient conditions, in the event of an accident-induced reactor shutdown, there is a sudden decrease in core power, the drive pumps come to a halt, and the outlet coolant temperature from the core sharply drops with a decrease in flow velocity. This prevents the coolant from entering the top of the hot pool for mixing and heat exchange with the high-temperature fluid. Due to the buoyancy effect, the coolant can only accumulate at the bottom of the hot pool, leading to the occurrence of thermal stratification within the pool. Considering the high heat transfer characteristics of metallic sodium in sodium-cooled fast reactors, the drastic temperature changes in the coolant during thermal stratification phenomena can be fully transmitted to the reactor components, resulting in equipment damage. The Computational Fluid Dynamics (CFD) method is a crucial tool for studying the thermal–hydraulic behavior of nuclear reactors. Its advantage lies in the ability to simulate three-dimensional turbulent flow fields and temperature distributions within the reactor vessel. Therefore, it is necessary to conduct CFD analysis for the analysis of stratified flow. The buoyancy term and the Prandtl number are critical factors influencing the accuracy of thermal stratification simulation results. Compared to fluids like water and air, liquid metals have a smaller Prandtl number, leading to the separation of velocity-temperature boundary layers. Under normal circumstances, using the Reynolds number assumption with a thickness equivalent to the velocity-temperature boundary layer can introduce significant computational errors. However, further research is needed to explore the correlation between buoyancy-dominated large-scale natural circulation flow and wall effects. A smaller Prandtl number in liquid metals results in a larger turbulent Prandtl number. Selecting an appropriate Prandtl number model is crucial for accurately simulating turbulent heat flux near the thermal stratification interface. This paper is based on an emergency shutdown experiment conducted in the upper plenum of the Monju prototype fast-breeder reactor in Japan at 40% rated power. It investigates the irrelevance of wall slip in this context. The buoyancy term is modeled using the Boussinesq approximation, accounting for the small thermal expansion in non-isothermal liquid metal flow. Simultaneously, the applicability of four different turbulent Prandtl number models is compared and validated. These models include the Aoki model, Kay model, Jischa model, and Cheng model. The results indicate that the presence or absence of wall slip has a minimal effect on the formation of thermal stratification and the upward movement of stratified interfaces. Furthermore, different turbulent models exhibit varying impacts on turbulent heat flux near the thermal stratification interface. These research findings provide valuable insights into the study of large-scale thermal stratification under accident conditions.