Warning: long post.
Just to make clear from my original post, I want to clarify that even if I powered OFF the camera body and removed the battery itself, upon powering back up, both of the Sigma 85 lens copies that I was testing caused the erratic timer-mode behavior on my 5D3. None of my Canon lenses exhibited this. Therefore, it is my belief there is something wrong with (certain) firmware revisions of Sigma lenses that, at least in this case, affects the 5D3 (in specific shooting scenarios).
FoCal has told me that they believe certain firmware versions (based on serial numbers of the lens) of the Sigma 85mm f/1.4 EX DG seem to have problems with FoCal while others do not. I'd be interested to know what the first three or four digits of the serial number of a lens is/was for those of you who have FoCal and have both had success or failures with this lens. The two that I tried and both failed miserably started with 129xxxx and 133xxxx.
Allow me to disclaim the following analysis as a "prosumer" of nine years, turning part-time pro photographer... so I don't know all there is to know by any means about our subject, but what I can apply is my engineering experience as lead technology architect for a Fortune 500 company.
There are a few key points mentioned in various replies that I want to address and also welcome for further comment from everyone:
1. Whether there is a need to "turn off" a Canon camera body when changing lenses.
2. Whether third party lenses, which have been reverse-engineered without licensed technology from Canon, are safe to use on a Canon camera body.
3. Whether FoCal has any part in this.
To start, regarding #1 - turning off a camera body when changing a lens.
Maybe full time Canon professional photogs here can tell me otherwise, but I have been using EOS DSLRs for nine years and have up to this point never turned off a camera body for the sole purpose of changing out a lens. In fact, I've never seen anyone else do it other than my overly cautious father.
Referring to pages 39 and 40 of the 5D3 manual, there is absolutely no mention about needing to do so, even in the foot notes. Researching further, one discovers that of the contacts on the body which mate to the lens, the contact which brings power to the lens from the camera body, is the last contact to "short." Thus, the lens is without power until the absolute final point of rotation when mounting a lens. Similarly, when removing a lens, it is the first contact to "open" removing power to/from the lens to prevent power from crossing into or out of the lens.
One should then consider whether turning off the camera should be done for installing/removing a hotshoe flash, or plugging in the USB cable, etc...
Additionally, have you noticed that when you turn OFF the camera body using the main "power" switch, since the 40D or so, there is the sensor cleaning that takes place? Therefore, the camera body still has power running through it and the power toggle switch is not a true "analog" power switch, fully removing power from the camera body, instead it is simply a digital action, disabling the controls and upon sensor cleaning completion, placing the camera into a very low power state.
Similarly, when you open the CF/SD compartment door, an interlock is broken and power to the camera is removed. In fact, I have seen this to completely remove power - nothing works whatsoever, because removing and inserting memory cards COULD lead to a short. They weren't designed the same as the lens barrel contacts.
If you truly want to be safe then, one would presumably have to remove the rechargeable battery pack from the battery compartment before changing a lens, or mounting a flash.
With respect to #2 - third party lenses being reverse-engineered.
It is well known that third parties have to reverse-engineer the autofocus interactions between a lens and the camera. Additionally, the lens mounts must be as well. They do not license these technologies from Canon. It is known also that because of this, future Canon camera bodies *may not* work well with previously released Sigma (or other third party lenses) because at the time of the lens design and manufacture, the third party was only able to reverse engineer up to that point with the Canon bodies that were in the marketplace at the time.
Canon on the other hand, can ensure backward compatibility with their previously released lenses in future bodies because they can place logic into the camera bodies' firmware to do so. They have no desire, need or motivation to do same for any third party lens.
After wanting so bad for the Sigma 85 to work for me (due to budget) and having had such an experience with two copies of it, it is my personal opinion that it could downright be more dangerous to mount a reverse-engineered third party lens to a camera body (even with turning the power switch OFF prior to changing lenses) than it is to change Canon OEM lenses with the power switch in the ON position.
For two Sigma lenses to cause such odd timer mode behavior within my 5D3 and cause FoCal to crash repeatedly, when none of my other Canon lenses (I have eight) ever did, is incriminating. I believe the two Sigma's I tested and their reverse-engineered "communications" with the 5D3 (in this case), caused the firmware/OS within the 5D3 to enter an unstable state.
I personally will never mount a non-Canon (OEM) lens to my camera body again.
With respect to #3 - FoCal's role in this
As a software engineer and I.T. professional of twenty years, I personally believe FoCal needs to clean up the reliability of their application. I do not believe FoCal is at fault here for anything the Sigma is/was doing to the camera's operation, but they could and need to do a better job... Simple things like performing proper error-handling techniques are not being employed within the application. For those with any software background reading this, they don't seem to bother catching and properly displaying to the user exceptions in certain scenarios while under test. Instead, one has to look into the Windows Application Event Log to find that nasty error messages (stack traces) when the application blows up. Instead of "catching" these exceptions (errors) and restoring a camera's settings, FoCal just blindly crashes because they haven't done their part within the code to "cleanly" exit from a bad situation. I believe FoCal crashing on the two Sigma lenses I was testing was due to a condition (mostly during LiveView target focusing) whereby the Sigma may not have found peak focus "in time" for FoCal to be satisfied. Instead, it threw FoCal into a tailspin nose-dive and FoCal has not been coded to accommodate this situation.
Just my two cents...