It's hard to tell without more information & EXIF data.
Here are some possible explanations (and why more info is helpful)
- If using "One Shot" focus mode, if a subject (or camera) moves after the camera locks focus then it will not re-focus.
- AF point review will indicate which AF was used to lock focus, but it is an indication of which AF point was used at the time of focus and would not account for subject or camera movement.
- Sometimes it isn't safe to photograph a particular subject with a shallow depth of field, and using an f-stop to slightly increase the depth of field is a good way to get little insurance. When I want a shallow depth for field and deliberately blurred background, I usually take a few shots at a few different f-stops so that just in case the low focal ratio shots don't give me the focus I want, I still have a few shots with a broader depth of field that I can use.
These are just a couple of examples where it might not be a lens issue. Of course it may be a lens issue after all.
When testing lenses for focus accuracy (or any other optical test), care should be used to ensure no other issues adversely impact the test. Whenever I purchase a new lens, I always test it while it's still within the return period.
Testing
A focus test chart would be ideal (such as a Datacolor Spyder Lenscal or a LensAlign testing tool. These tools have a flat card with a focus target. Adjacent to the focus target is a scale positioned on an angle -- which extends both in front and behind the focus target. You focus the camera on the focus target and take a photograph, then inspect the scale to determine if the most accurate focus is at the target vs. closer or farther.
When I use these tools, I deliberately de-focus the camera closer than the target so the camera is forced to focus outward to the target. I repeat this at least 10 times. Then I focus the camera farther than the target so that the camera is forced to bring the focus forward to the target and again, repeat at least 10 times (you want to make sure you have a good sample of tests just in case one or two tests are outliers from typical behavior.)
The cameras you are using all support AFMA (auto-focus micro-adjustment). If you are able to identify a consistent trend you can apply some AFMA and re-test until you are satisfied with the results.
The camera should be on a tripod and the focus target should be on a stable surface (don't have someone hand-hold the target).
Use a remote shutter release to trigger the camera so that the test isn't affected by any vibrations caused by touching the camera.
Using a commercially designed test target isn't a requirement -- the real requirement is that you try to isolate everything else that could impact focus (photographer movement, subject movement, vibrations, etc.). But a commercial test target isn't particularly expensive and when you're buying top-end glass ... why not make sure you've got it dialed in as well as possible.
