It seems pretty unlikely that 100% of the people on the outbound train would use the app and trust the historical data to go to the platform early.
The platforms are very narrow for the LIRR tracks, so holding 200% of a train (outbound and inbound) would be sheer pandemonium, and would practically guarantee someone falls off the platform. Even holding 100% of a train (just inbound, disembarking passengers) is already straining the platforms and methods of egress to their limits.
Holding even just 110-120% of a train load is very much a safety issue. I'm with the MTA on this one - the "mad dash" is horrifyingly inefficient, but is the safest course of action.
Of course, the correct fix to this is to fix the platforms such that passengers can wait at track-level without safety issue. But, of course, that's a multi-billion dollar problem nobody seems willing to touch.
I'd hate to imagine what would happen during peak if a train came in on an unexpected track AND a different train on the expected track - it is not unusual for multiple trains to be announced at once. Then you'd have a pile of people trying to go up those narrow stairs while a pile of people were trying to go down.
As somebody who commuted for a long time at peak rush hour back and forth from Penn Station to D.C., I can tell you that following people who "seem to know where they're going" would not be a reliable strategy.