What many has it been confirmed by? So far I have seen one person say that he found a difference with the lens cap off and it ONLY occured at EV1, the absolute bottom limit of the metering performance range, where it underexposed by 1/3 of a stop.
That sounds like me you're referring to, and you're making the situation sound significantly worse than what I discovered.
Set the camera in manual mode to EV 1 (1" @ f/1.4 @ ISO 100), and find a dark corner of a dark room to point to until the meter bug is centered. Turning the LCD backlight on and off will not budge the meter at all. This is the bottom of the specified range of the 5DIII meter.
Set it to EV 0 and the backlight still won't change the exposure.
At EV -1, the backlight moves to the right by 1/3 stop when you turn it on...but you can't say that it's causing underexposure at that point because it's operating outside of its design parameters and may well already be off by some unknown and unspecified amount in either direction.
At EV -2, the backlight still only makes the meter move 1/3 stop to the right.
At EV -3, the backlight makes the meter bounce between 1/3 and 2/3 stops.
At EV -4, the backlight makes the meter move solidly on 2/3 stops to the right...and you're five stops below the bottom of the meter's range (30" @ f/1.4 @ ISO 100) and basically in the dark. There's no way the meter's readings are reliable, with or without the backlight on. Any photographic meter -- not just the one built in to the camera.
If you're actually photographing in conditions even as dark as EV 1...well, either you're doing long-exposure tripod work, maybe of the nighttime skyline, and you know full well that your meter is already useless...or you're being silly showing off your new toy outside at night away from the lights, and probably shooting at ISO 25,600 or something ludicrous and you know the results, though surprisingly not bad, will be totally useless for anything serious. And at EV -2 when the "problem" actually starts to appear, and only at a third of a stop? We're talking moonlit landscape photography, where no sane person even thinks to have a look at the meter. Hell, you're probably thinking of shooting in bulb mode with a stopwatch in those conditions....
I think it was someone else since they mentioned EV1 not EV -1:
"Since the operating range of the 5D3's light meter is Ev 1 to 20 according the manual, I decided to try testing for the light leak phenomenon at a light level equivalent to Ev 1--the minimum light light level at which the 5D3's light meter can be relied upon to provide an accurate reading. Although I'm not an expert on Ev levels, after consulting a chart on Wikipedia, I determined that at f/5.6 and ISO 100, Ev 1 would correspond to a shutter speed of 15 seconds. (If I'm wrong about this, I'm sure someone will tell me.) So I set my 5D3 to ISO 100, and my lens to f/5.6, took off the lens cap, covered the viewfinder eyepiece, and found a dark environment where the 5D3's meter gave me a shutter speed of 15 seconds. Then I pressed the LCD illumination button to see what, if anything, would happen.
Having previously convinced myself that the leak leak phenomenon was only likely to manifest itself at Ev levels far below Ev 1, I fully expected to see no change in the shutter speed. But to my surprise, I did see a change, although only a slight one--from 15 seconds to 13 seconds. Shining a flash light directly into the top LCD resulted in a similar change.
I then repeated the test by moving to a very slightly brighter area where the meter gave me a reading of 5 seconds--corresponding to something just over Ev 2. When I pressed the LCD illumination button this time, I saw no change, nor did I see any change with the flashlight.
So, what conclusions do I draw from this. Well, contrary to my earlier assumption, it does appear that this phenomenon could affect metering within the 5D3's published metering range--but only at the very bottom of the range--around Ev 1--and only very slightly--the difference between 15 seconds and 13 seconds. If you're trying to meter at lower light levels, say around Ev -4 where most of the testing I've seen posted has been occurring, the effect on the meter reading will likely be more significant, but of course, any metering you might try to do at Ev -4, or any other level below Ev 1, is already inherently unreliable."
In either case though it doesn't seem like a problem to me and certainly nothing to be embarrassed about, come on.