Here is a direct comparison between the two modes for reference.
The difference also scales as light count goes up, so potentially Forward+ can be vastly faster.
The only difference in the code is where lights are looped. Forward plus uses the specific iterator, while in Deferred i pass the light count from the light culling result and loop with that.
Not sure how is all handled internally though. I will do some tests with iterating the lights in Forward+ using the culling result to see if that is the only reason Forward+ is so much faster.