Hi, I agree with all the others, that I’d more quickly fault a t3 than the lens, but I think it would be disappointing to upgrade the camera and realize that it was the lens after all.
Here’s the problem: you say that it hasn’t always done this. That’s odd, for an afma problem. If the combo was working perfectly, and then lately you pull it out and it doesn’t it indicates that the state of your equipment has changed. It can happen that shutter boxes and mirror assemblies get messed up, but it would seem far more likely that the lens would get messed up. (much more complex, many more moving parts, etc)
@jamesstaddon, I agree about being more willing to credit a t3 with a mistake than an L-lens, but in the case of the 100-400 I’m less sure. I researched this lens quite a bit in the last year and basically read that there is a lot of sample variation and a lot of them need afma. (60d doesn’t afma either though the 50d and the 70d do) Because of all the reviews I read about it, and for other reasons, we went with the 70-200 and 2xIII, though it was a rather more costly route.
@thefarmhand. True about full-frame, but the options are tricky. The 6d and 5d II have basically the same autofocus system as a rebel. (11 and 9 autofocus points respectively with one cross-type in the middle) So your full frame options with a better focusing system are in the 1d series or the 5d mark iii and both options are somewhat more costly. If you’re not printing bigger than 8×12 the 1d iii is interesting I suppose.
A new body might fix focusing errors. It would at least give you afma if that were indeed the issue. I might add that the 7d mark ii has apparently had some autofocus issues itself, and canon has been servicing a lot of units. In your shoes I’d be looking hard at a used 1d iv for it’s faster lens drive and autofocus-linked spot metering, if you’re in that price range. Otherwise I’d look at the 70d. A bit less and I’d look at the original 7d or the 1d iii, used.
I personally would be afraid of buying a new camera and finding out that it was the lens after all. Do you have any other lenses that would give a shallow depth of field? Test them out and see if the autofocus accuracy is better, or the same. (don’t bother with testing third-party lenses, such as sigma tamron and the rest, autofocus accuracy is always a wash with non-proprietary lenses) If your other lenses perform perfectly, all fingers pointed at the 100-400 as the culprit. If they all perform similarly, your t3 is probably guilty.
I would also send the combo through some testing: set up a high-contrast test target and your camera on a tripod. Also place something such as a ruler beside the target at an angle so you can see it and so that the ends will be closer to the camera and further from the camera than the target. Then focus the camera, then unfocus the lens manually, focus again, unfocus manually again, examining every picture. You can look at the ruler to see where it was focused compared to your target. You might start to see a pattern of off-focus.
A back-focused or front-focused lens would actually account for the situation fairly well imo: A camera’s autofocus system is never 100% perfect. In fact, it is impossible to be “perfectly” perfect because, in theory the depth of field of perfect focus is actually infinitely thin. So the autofocus sends the lens to the “right” place within given tolerances. Lenses and cameras are also built to given tolerances. Supposing that the the tolerances are an arbitrary +/- 3 focus units, a camera of, say, +2 with a lens of -2 will seem perfect, even though neither lens nor camera is perfect. The same camera at +2 with a lens that is dead on at 0 will only seem so-so even though the lens is actually closer than the first. The same camera at +2 paired with a lens at +2 will seem actually pretty bad even though neither are really beyond tolerances. With an 18-55 kit lens, it would be impossible to detect these discrepancies, however with a longer lens, or a fast prime, such as the 85mm 1.2 the depth of field becomes so small that any errors are immediately apparent. Plus the buyer of a 2000$ lens is 10x more committed to having a perfect lens than the buyer of a 200$ kit lens.
Now, autofocus systems being made by imperfect people are also imperfect, especially our phase-detection autofocus. (As a side note, if you were to always use your live view, contrast detection autofocus, your in-focus rate will almost certainly rise) Sometimes they miss. Now, if your camera-lens combo is, say +4, your autofocus will normally get it wrong. But, if your autofocus misses and arbitrarily lands on -4 the image will be tack-sharp. So what I’m saying is that, if this were the case, the times your pictures end up sharp would be your camera-lens combo actually getting it wrong.
Read more on this by a person more knowledgable than me here:http://www.canonrumors.com/tech-articles/this-lens-is-soft-and-other-myths/