I would say
Double, triple, quadruple,...
track it (whatever works with the singer voice) adjusting volumes and pan so that it seems like only one track (one is louder and centered and the others have lower volumes and different pan if it works), performances have to be very tight to each other.
Compress it with a 1176 very hard with the fastest release you can get without pumping and the fastest attack without taking too much transient energy out so that it doesn't lose the strength.
Eq it to make the words or screams
more intelligible and so that it fits the rest of the instruments (having the music with a darker sound and the vocals with a brighter sound doesn't work very good or the inverse IMO)
Distort 1 or more of the lower volume tracks to give a fuller sound. (It helps to bridge words better if the singer doesn't have a good natural voice for it)
Add delays 1/4 and 1/8 and a touch of reverb but without noticing it when the instruments are playing.
In the end how I see it is that when you have all the processing on you shouldn't hear it separately, but when you take the processing out you can definitely miss the processing and hear the voice weaker.