Discussion started in Youzhny def Nadal in Dubai thread, about the ball that was overruled by Hawkeye and won Youzhny the first set (7-5 in tb). I think the subject is worthy of a separate thread.
As the Hawkeye has a margin of error of 3.6 mm (source unknown to me), it can't make 100% safe calls less than this. Apemant came up with an alternative solution: let Hawkeye have three alternatives IN/OUT/UNKNOWN, and replay the point if the UNKNOWN call occurs. Maybe more fair than today's rules.
The problem with this approach is that it can't be used for ordinary line calling without significantly delaying play. Lets assume that an average line judge as a margin for error about 5 cm. Judging from a diagram I found on the web
it seems as if there were about 30 1st serves called "in" from Sampras-Agassi TMC final 1999 that landed so close to the lines that they couldn't have been safely called by a human eye.
To replay all these points would have caused significant delay, and there probably were a lot of serves called "out" that would also have caused replay, in addition to that a number of ordinary rally points.
One could of course use the unknown option only when Hawkeye is at hand, since a lot fewer balls would be uncertain, but it would make little sense IMO to use it when the calls are more accurate.
One way to solve it would be to say: if the ball according to measurement techniques were LIKELY in, then it is in. (The ball that started this discussion was more likely in than not.)
That is the present paradigm, and IMO it is enough. Particularly considering that getting favourable line calls is probably roughly equally likely for every player.