Transient noise is just as accurate as you specify it.
You need to specify a very high frequeny range, which leads to small simulation time steps.
If you want to measure high and low frequency noise effects you need to simulate a pretty long period of time.
Both of this together lead to the classical problems of time and memory

My own experience is that behavioral models with modeled noise effects are pretty accurate compared to their parameters, but transient noise analysis of transistor level circuits often differ in their results.
Or like Ken said in one of his papers - "transient noise analysis is very error prone."