A tl:dr provided in first comment.
As a medic main in 6s I have never had a satisfactory answer for this question. People parrot the wiki or they say their own side but none of them felt correct to my own feelings, so I decided to sit and time many ubers to give my own "data-backed" answer since I honestly did not know the answer to the question. and instead relied on the feeling that each flash reduced uber by an equal amount, independant of who got flashed.
Overall the intuitive answers I received from others as well as my own understanding worked well in judging how it worked for clean gameplay purposes. The fading invulnerablity of uber or what most people know as the "flash", being a portion of the uber shared to multiple people so it fades faster is an understood point, and counting how often that happens is what most people intuitively do to determine a better or worse uber.
The wiki under the "activation" section reads:
With a single patient, an activated charge lasts 8 seconds, draining the meter at a rate of 12.5 points per second. The Medic is free to move his healing beam to another player, which will give the effects of the charge to the new patient, but the total ÜberCharge duration decreases by half a second for every second that another player is being ÜberCharged.
This line, along with the following table are what players tend to use to determine how flash timings work:
https://i.gyazo.com/821e71fa9af0dd09041464daceb5b834.png
Personally I understand this to mean two things:
1. When you flash a player you lose 0.5 seconds of uber from the 8 seconds.
2. This occurs when you flash any player whether they have been ubered already or not.
We reach the conclusion of #2 by extrapolating the table timings.
0 recipients = Solo uber on medic of 8 seconds
1 recipient = Solo uber on one player of 8 seconds.
2 recipients = Both players consistently ubered for 5.33 seconds. If you only flashed once, then the uber length would be 7.5s as only half a second is lost based on the previous line of information. So this means that this timing is for both players to be constantly ubered.
3 recipients = All 3 players consistently ubered for 4 seconds.
etc.
This is how I understood the wiki, but due to it's lack of explanation it is confusing. The wiki does not explicitly state if the recipients are the same people being flashed or not. In other words, does ubering 2 players constantly in 7 flashes result in a same length uber as 3 players in 7 flashes? Is each "recipient" an individual flash? My understanding could be misunderstood and my timings could be off massively by a few seconds every uber.
The wiki also lacks a lot of critical information to actually help players to understand how to flash players. How long is a flash? Does the full duration of a flash need to occur to drain the extra half second of uber? What is a recipient? When does the extra half second of drain occur?
I've sat and done a test of multiple ubers in order to determine the answer to some of these. I classified a "recipient" of the uber to be someone who recieved the uber at any point (ignoring the medic who used uber). I classified a "flash" as a break in the beam. A regular flash waiting for the uber to fade before reflashing (if applicable) and a "quick flash" as an instant re-flash (if applicable).
My data is rough and there are more tests than just these ones but these were all that were recorded. Time is from pressing M2 (bound to timer start) to when I see a 0% on my hud (high possibility of human error):
https://i.gyazo.com/ae2fcdf8251978bddaccff9abf4642c7.png
The top table represents regular flashes. I tried to keep the full flash before rehealing the same person, as in the uber flash had faded before they got ubered again. This resulted in fairly expected results based on the wiki's information of 0.5 seconds of drain per flash.
The second table shows quick flashes. These are ones where I would spam click to get the flash as quick as possible onto the next (or same) person. This became a necessary thing to test as my initial timings were not nearly as consistent as I would have liked, and many results were completely innacurate when testing, which I now know was due to re-flashing too quickly rather than counting errors due to the similarity of results between the two tables.
Based on these results and my testing, I determined that:
1. The wiki timings are likely correct for my understandings of them.
2. A flash is 1 second long (requires further testing) from the moment the heal beam disconnects to the point of vulnerability.
3. If this full second does not occur then less of the 0.5s drain will occur. This is likely linear (e.g. if you reflashed after 0.2s then the uber would only go down by 0.1s total to result in a 7.9s uber) but the exact relationship cannot be stated. In other words, the rate of uber consumption increases upon a flash rather than a chunk of 0.5s being taken from the uber.
4. The increase drain rate occurs at the disconnection of the heal beam. This means that if you uber without a target and then start healing someone it will not reduce the duration of the uber the same as a flash would. This is why a flash must be known as a break in the heal beam rather than the gifting of a player invincibility.
Some examples:
If you consider a flash as to how most of the community does, you will see the problems this causes when counting uber durations. You can notice that a player flashing everyone once but then holding uber on the last individual will have flashed 5 people, 4 times for about 6 seconds of uber duration, whereas another individual flashing 2 players, 4 times between them will also have this duration, if not less due to reflashing the same player, making ubers incredibly deceiving. However, if you ignore the number of people ubered and instead only focus on the breaks in the heal beam, you notice that both teams have 4 breaks in the beam, resulting in an even length uber trade.
This is also important for medics to understand when to back up from an uber. If you take a solo uber and then the person you are solo-ing jumps away 1s before the uber ends, the uber will finish and start to fade within 0.5s shortening the time you thought you had left exactly the same as if you flashed a new player. This also means that any breaks in the beam during a solo decrease the duration of the solo uber. If the player being solo'd at any point was vulnerable, then it would be the same as flashing a brand new player into the uber.
The final point to note is that an uber spam flashed between two individuals 20 times will last the same duration as an uber carefully flashed 5 times to keep both members ubered for the 5.33s time. This is a logical step from point #3 above in which the rate of consumption goes back to normal if the player had the uber flash on them is now in the uber again.
Ubers are difficult to get right, and a team that can use uber trades to their advantage will be head and shoulders above one that can't. For example, if you wanted a solo scout trade into the enemy team, who should the scout shoot and who should the opposing medic flash? If you go too far in to force a flash it is possible your medic will be disconnected and therefore force a flash of their own, resulting in a same length trade, even though you forced a flash. The scout running in should also not be concerned with doing damage towards the end of the enemy uber, as then they will in essence receive a "free" flash as the last 0.5s of ubercharge can be used to flash another teammate for 1s.