Sometimes these things are a total mystery. I had one with Eos 3 film bodies and the 85mm f1.8 EF. Neither of the two bodies I had in Afghanistan would focus accurately with either 85mm f1.8 I bought. I tried one lens, got a replacement and it was the same. What's odd is that focus was perfect through the viewfinder, but miles out on film as distance grew. The greater the focus distance, the greater the error and by 5m it was about 3m out! At minimum focus, it was fairly accurate. At, say, 3-5m the error was so huge that even stopping right down could not account for the massive error. I had used both Eos3 bodies on lenses from 24mm to 200mm and they were performing perfectly in all cases. In all my years I had never experienced anything like this.
At this point I could not understand how both 85mm f1.8 lenses could do the very same thing on BOTH my Eos 3 bodies and the error occurring defied reason. How could the image be perfect on screen, the camera lock focus perfectly at all distances and the actual negatives look like I had focused in a different county?!.... yet when I changed to ANY other lens I owned, everything was peachy perfect. It all pointed to the lenses, but how? What?
In desperation I ordered a 85mm f1.2 L II and that lens was perfect on both Eos 3 bodies. It later proved to be just fine on the Eos 1nHS body I had (but not with me at the time I was trying the 85mm f1.8 lenses). Years later I bought a third 85mm f1.8 for my 5D III and it too is perfect on my digital bodies.
I never did figure this one out and nobody has ever been able to explain it to me either. I went through about ten careful test rolls trying to find something to make sense of and it drove me nuts. Here I am bringing it up seven years later! It makes no more or less sense than your problem with the 5D Mk IV!
I got my portraits in the end (you can see some here:
http://www.thomasstanworth.com/afghan-heroin-not-for-export/) and I absolutely love the 85 f1.2 L II as a result. Maybe it was meant to be
