I think all this talk of "realism" in the game is really about "consistency" and not really about being "realistic" in comparison to our real-world (although many characteristics will likely overlap). The DM needs to present a consistent game world where players aren't left guessing if their abilities will work at any given time. And, when something doesn't work as expected, there needs to be a strong reason within the game world that is both discoverable and communicated. In the case of Sentinel not working against a particular creature, that needs to be explicitly telegraphed ahead of time, IMO. And, if not, then the DM needs to lean into being a human who is capable of making mistakes, give the player the information their PC would/could/should have known, and let the player decide what to have their PC do instead. That's not to say that an occasional "combat puzzle" isn't okay - but it should come front-loaded with clues rather than some kind of trial and error experiment where the player has no way of knowing what will or won't work, leaving it basically up to chance or guesswork. Granted, they may have missed most/all the clues but at least the clues will make some sense in retrospect.
Now, at our table, Sentinel would work vs a kaiju. There are plenty of in-fiction reasons we can come up with, many described upthread, why it can be so. And it won't ruin my pre-determined DM story (namely b/c I've learned to let go of that concept many years ago and just let the story tell itself through the contributions of all at the table); nor will it ruin verisimilitude since we have all bought in that this is a fantasy world with magic and powers that don't map directly to IRL. That's just how we play. At other tables, of course, that mileage may vary.