If the new solution requires that each of the three models computes a prediction for every record, the time efficiency during inference will be reduced. This is because the inference process now involves running multiple models instead of a single model, thereby increasing the overall computation time for each record.

In scenarios where inference must be done by multiple models for each record, the latency accumulates, making the process less time efficient compared to using a single model.

References:

Model Ensemble Techniques