In my case I work with the "Registration Memory" by songs, by style and general registers, where I leave the main voices and styles at hand if needed. So I use the registers in 3 different ways.
Some songs use up to 4 or more voices, so RM will have to be per song. Here in Brazil, we use the same style to play for about 30 minutes, 1 hour, 2 hours (Ex.: Forró, Bossa Nova, Samba, Pagode, Jovem Guarda, etc...). So we just changed the voice and the style remains the same. In this case it is more interesting to make an RM with 8 different voices and leaving the original style saved in all of them.
Another example would be when I use the RM to play a "Pout-Pourri", with 3 or 4 songs in the same style, but with different voices. In this case the RM will be prioritized by style.
I also use dozens of songs in a single RM, where I put the name of Singer in the Bank, and each RM has a different song (for example: Bank Eric Clapton1: RM1 - Tears in heaven, RM2 - Something, RM3 - Change the world, etc...)
There is no better or worse way, but what is the need.
I hope you understand what I want to give you!
Enildo