Higher frequency response, see also handbook page 105, but tbh, do you really require it ?
Your PC has to process the double amount of data, is the PC capable to do so reliably ?

Side notes:

If you have other digital sources they also need to be capable delivering 88.2/96, as you can not run different clocks without SRC.

For what reason would you require lower ADDA latency in a life context below 0,9ms ? If I didn't calculate wrong thats a latency as if you stand 30,9cm in front of the speaker.
In a life context your audience can be i.e. 5-100m wide from speakers, which is a latency range of 14 - 291m.
Who cares then about 0,9 or 0,3ms ?

Yes AFAIK latency in standalone mode is lower at higher sample rates, same latency in samples but half the time in ms. But latency is very very low at 44k too. In a live situation on a stage I would expect the latency because of speaker distance and the speed of sound to have more impact ;-)

That is more a subject for discussion then a fact. 96k will allow higher frqs then 20khz, but we can't hear them, I am 57 and can hear up to 15khz. Some people claim the can hear or feel the the higher frequencies. You can always try it. If you like how it sounds, great. But my guess is that if you do a double blind test, you won't know which is 48k and which 96k.