tbh I was doing upwards of 25% of King of Chaos' HP with Tenshi too. Although, I think something might be going wrong there, because when I tried reverse calculating the damage done from my recorded run to figure out what kind of DEF King of Chaos had, somehow Tenshi was doing more damage than possible even with 0 DEF. I figured out with further testing that King of Chaos has about 4m DEF, but that just raises more questions about why Tenshi was hitting such high damage numbers. Under the most extreme circumstances (SPI weakness being 50, maximum damage variation of 1.1x), Tenshi's still hitting about 33% harder than she should by my math. World Creation Press was also doing similarly higher amounts of damage than it should've been doing. At first, I thought that Sanae's Youkai Buster was bugged and applied to Oni type enemies (Super Youkai Buster is 1.2 * 1.16 for 39% more damage, which would explain the discrepancy), but from testing it further, this turned out to not be the case. I've no idea what it could be that caused this. It's possible that if I tried testing it for other bosses, I would discover the numbers are higher than they ought to be as well. What I can confirm is that Tenshi's damage was normal on the King of Chaos when testing Direct Attack with no buffs outside of Girl of Bhava-agra, Strategist, and some small % DEF/MND for Courageous Sword, so those things aren't broken.
But anyway, that's reassuring that the original postgame was similarly imbalanced, since it means that the Plus Disk bosses also have a strong chance of being rebalanced. I won't worry about it then.
EDIT: Ah wait a second, I was testing Direct Attack with Murakumo's Blessing on, but that was before confirming that Murakumo's Blessing buffed Iku's Attack command, and by extension, should buff all Attack commands. I didn't account for that, which means King of Chaos' DEF should be even higher than 4m? That makes no sense though, that can't be right. I don't get it, the damage calculations I did were accurate prior to the 1.104 update.