Even if a model had dropout layers, Model Optimizer would handle them just fine. In the case of TensorFlow models, Model Optimizer would expect a frozen TensorFlow protobuf as input. The online document below is detailed documentation regarding how Model Optimizer performs its optimizations. Note that since dropout is a training technique which helps to resolve the problem of overfitting, from Model Optimizer's perspective, it doesn't care. Its algorithm removes any and all bits related to training anyway since OpenVino is all about inference, not training.
Hope it helps.