I have the exact same issue with my 5d3 as RobertWelch. I used it at two events, with two of my other cameras, a 1Ds3 and a 5d2. I set the time on all three cameras to within half a second before every event. I use 600X Lexar CF cards, format them in camera before the event, and shoot in raw only. I switch between my three cameras very quickly, often taking photos within a couple seconds after switching. I imported all the photos into Lightroom and sorted by capture time. I noticed that the 5d3 images were sometimes out of order.
Prestonpalmer reports inaccurate clocks that drift slightly. I concur. But I've used three cameras simultaneously at events for years, and have never had a problem with image order as I've had at the past two events with my new 5d3. I synchronize all the clocks before every day's event, and they usually stay fairly accurate within that day, even the 5d3. When I noticed the problem, I started checking the clocks after each event, and have found them all to be within a second of each other. So clock drift is not the issue here, at least not for me.
I conducted some tests, setting the 5d3 clock to my watch, then photographing the watch and comparing it to the encoded time. I found that if I shoot right after setting the time, or after reviewing a previous image on the screen, the time encodes correctly, right to the second. But if I shoot from a blank screen (as I normally do), the encoded time starts to drift, anywhere from a few seconds to half a minute EARLIER than the time displayed on the watch. It's not consistent, but the encoded time is always earlier. I've seen as much as a 36 second difference. I took about 40 photos in one test, and all of the shots taken right after visiting the time screen or reviewing a previous photo were encoded correctly. But almost all of the photos taken after that were off by varying amounts.
I suspect that this is a software problem, and that the encoding algorithm is not referring to the clock for each image, but somehow keeping its own time, inaccurately. Then occasionally updating its time from the clock.
I think this is a problem that typically only an event photographer would notice. When using a single camera, who's going to notice that the encoded time is slightly off (unless you are a clock aficionado)? Therefore I wonder how prevalent, yet unnoticed and under-reported, this problem is on the 5d3.