if you want to keep the layers 3d and be able to control them in the main composition
you would need to turn collapse on and the 3d switch as well (which is what you do 99% of the time
with groups of 3d layers that need to be pre-composed) and yes you would "lose" the original coordinates
which makes sense since it is now a new group with a new anchor point and position.
parenting in this scenario would seem the preferable option. it is basically a way for grouping layers
and with the advantage of staying in the same comp.
leave transforms in composition makes sense that its not enabled for a group since a group has a new coordinates and values. if you select more than one layer how would AE know which values to keep? by average? doesn't makes sense that it would work any other way than it does to my opinion.