I get the feeling you think collapse transformation means something that it doesn't.
What Collapse Transforms does is to leave all the transformation properties in the subcomp intact and available to the comp that it's nested in. It's basically like you've got all the layers still in the outside comp, just grouped into one layer for ease of organisation.
Without collapse transforms on the subcomp is pre-rendered and treated like a single piece of footage. So, for example if any of the layers of the sub-comp extend off the edge of the sub-comp's boundaries they won't be cropped with collapse transforms on.
Here are two subcomps, both contain text that extends past their outside edges:
And here they are in the outer comp. Top one is collapsed, bottom aint.
If you scale down a raster layer in the subcomp and then scale it up to its original size in the outer comp, without collapse transforms on it will be pixelated, because it gets pre-rendered to the resolution of the subcomp, and then scaled up in the outer comp.
Here they are scaled down in the subcomp:
And here's the outer comp with both layers scaled back up:
Another effect of collapsing transforms is that you lose the ability to change the transfer mode of the subcomp in the outer comp. The layers in a collapsed transform-ed subcomp apply whatever transfer mode they have inside the subcomp, again, like they were actually in the outer comp, just grouped together.
What Collapse transforms doesn't do is automatically crop the subcomp to fit its components. Is that what you were expecting?