I can’t give you the exact proper technical answer as I’m not that experienced but my feeling says that this is dependant on the fact that both the LUT itself can have limitations depending on what it does and that a display space like rec.709 is not exactly reversible to scene referred data.
On top of that, in using ACES you are using double tonemapping now. The one from the lut (then back) and then again from ACES. It will still inherit the ‘ACES Look’ which the lut by itself doesn’t.
What you want is the ARRI look emulated in ACES space without the display mapping.
Have a look at this thread, I think this will get you in the right direction.