They vary because ERA is a bad indicator which is prone to BABIP fluctuations (hence the variations), which usually ends with it regressing to FIP over a large enough sample (usually doesn't take very long)
So all of this fluctuation you're seeing is because ERA is the bad stat, not FIP. Because batted balls are prone to randomness whereas peripherals are much more talent-based and won't show as big fluctuations over the course of the year
There are certain player profiles that can be identified after at the very least several hundred innings, that might have a FIP-beating effect (mostly guys with high FB% that get really high exit angles), but this is just for your general knowledge, Hoffman isn't one of those guys
Don't know why you bolded the season's worth of innings part but should be noted ERA is still less effective in one game's performance. Pitch level statistics like TIPS from JFaS are best for single game but obviously one game really doesn't mean much