Tested 3 copies of this lens the other day at work. Only 3 copies, but extremely narrow variance. On par with the 50L (at 10 copies) even. Much better than the 50/1.4. The lens feels a lot nicer too, and internally is actually a lot more robust than the 50/1.4.
There was a time when Canon's QC was pretty crappy but they improved massively over the years.
The IS lenses are clearly worse though.
If you are ready for a shock, you should repeat the excercise with Fuji, Pentax, Sony (aps-c) or Samsung lenses... ;-)
We are actively testing many lenses and can test on most mounts, but sony and M4/3 present very real challenges with regard to their AF, Zoom, and OS functions. Essentially the new electromagnetic focus systems are loose when not powered, so reliable testing cannot be done without a camera attached - we are working on a solution, but the june rush is quite horrible. The 5Ds and 5Dsr also came in yesterday and we get 2 days exclusivity with them in the testing and repair department before they go into the fleet. The camera does very well and I am happy to say that the canon 300mm f/2.8 II, Zeiss Otus 85, Zeiss 21mm/2.8 ZE, Canon 70-200 IS II and f4 IS and the 135/2 ZE are simply brilliant on it.
I'm working on automating a large portion of the MTF testing we do, e.g those graphs. As of now it's as simple as opening the raw data from the bench, then Matlab will drop that lens copy into its folder. After many samples are collected a different program simply has you open the target average spreadsheet and then averages all copies of the lens, finally a third program makes some special plots of the lens and its expected performance.
We will be debuting a new score system in place of the old reliability reports, as well. Essentially last year a reliability report caused a headache since there had been about 3,000 in house repairs. By the end of the month we are estimating to pass 17,000 and they are not simply tracked in an excel sheet anymore, so a professional data analysis must be hired out to crunch the numbers.
Essentially the new system will rank each lens model for variance based on 10+ copies of each model, and the compiled data will grow over time to improve accuracy - but there are over 1,000 lenses to be tested to establish an initial database... preliminary results while I continue to evolve the software are that the nikon 24mm f/1.4G is over 2x worse in this respect than even the 50/1.8 STM. I fear for when we look into sony and samyang lenses.