Most would agree that grailing is most optimal for those servants that you enjoy playing + are effective gameplay wise.
If we’re strictly speaking gameplay wise, SRs are the way to go in terms of returns for your grails as many of them can reach SSR stats with 2 grails + gold fous. You can have 5 SRs with SSR stats instead of having 2 lv 100 SSRs. Extra damage you could’ve gotten from 5 grails can be sourced from other mechanics such as CEs and supports. Meanwhile, what doesn’t survive at 15k hp most likely won’t live long even at 16k hp.
In the end, its a matter of priciple. I, for one, gave grails to low star servants such as Bunyan, Arash and Spart (1 grail each) as increasing their damage widens the range of their farming capabilities/supporting capabilities. Arash, for example goes from 20k+ to 30k+ neutral damage and 40k+ to 60k+ effective damage (with max silver fous) which allows a more consistent gameplay result. On the other hand, I’ll never grail something I’ll never use even if I worship the servant just cause I don’t want useless trophies. Raising any servant from 90 to lvl 100 is a sore spot to me as it only increases the servant’s effectivity and not their viability. Most servants utility are the same whether they’re 90 or 100.
Lastly, what servants you have can contribute to your decision. Grailing my Hans in itself to 80 may sound to result in a mediocre survivability to most. But having something like Tamamo, his survivability goes from mediocre to great. His longer staying power magnifies his effectivity with his coming buff.
TLDR: In my opinion, go with Penth over Raikou if you’re looking for minimal grail investment (at least until 90). Also Raikou’s stats are sufficient enough for her role which is farming. Her bossing’s not that great until her crit buff. Even then, she gets outclassed by Junao.
In the end, agree with most to grail who you like but at least make sure you’re going to use them. The only time you disregard the gameplay part is if you only want to look at a pretty number such as 100.
Sorry for the long post.