It gets better. Stay with me.
We know that Nowi and F!Grima wouldn't survive the attack, but they were never going to survive the attack. Let's take a look at M!Grima and Tiki.
View attachment 143338
In order for M!Grima to survive the attack, I'd have to boost his Res to 35. Right now I can get it to 39.
Starts at 29 base. Boosted by summoner support, that's 31. Park Nowi near him, that's 33. Have Nowi use Rally, that's 36. That alone would leave him with 1 HP, and he'd drop Julia on the return swing.
With Tiki in range, that makes up the 3 Res Nowi would need to bring with her Rally. Positioned properly, she can then advance on one of Julia's teammates.
Now let's take Tiki.
View attachment 143339
In order for Tiki to survive, I'd have to boost her Res to 28. Right now I can get it to 30.
Starts at 21 base. F!Grima support, that's 23. Ward Dragons from M!Grima, that's 27. Rally from Nowi, that's 30. Drops Julia on the return swing.
Needs every bit of everything my team can squeeze out, but very doable. Would be better if I had Warding Breath fodder or a few more merges, but the resources are what the resources are.
So as it stands, I have already existing options to survive
that particular Julia. Heck, I even ignored the blessings.
And yes, you can argue that as a defender, the AI would never consider the Dragon Ball. But attacking my team with Julia, you've got three dragons I can't reasonably build to withstand her without Dragon Balling. Bait any of the vulnerable ones and the whole team falls apart. And Julia's supported by three other teammates, only one of whom has to counter Tiki if Julia somehow doesn't already (what up Reinhardt).
Attacking into Julia, meanwhile, has never been a problem. My anti-Julia strategy has been to close distance and kill her off before any dumb **** happens, abusing F!Grima's Reposition and a two-ranger like Tiki or Nowi, and it's worked every last time.
To be fair, by the same token, I could run Warding Breath on Tiki and tip them most of the way back my way. You'd come out on top by one reduced point of Res, but my above set had that in its margin of error.