Under normal circumstances, this is true. Unless you use G92, your Z should be persistent.
I can think of a few exceptions:
- bit shifted inside collet (collet nut not tight)
- router shifted inside mount (mount not tight)
- workpiece shifted up (possible with side clamping e.g. cam style)
- X axis tilted up about the side plates (possible in a Z crash)
.5" is hard to imagine though. It would make me suspicious of lost steps.
EDIT:
Which reminded me of this.
This was a very weird situation. I once experimented with 2:1 gearing and 1:1 ustepping on the Z (as opposed to 1:1 gearing and 1:2 ustepping) and experienced consistently lost steps that were consistently affected based on acceleration and junction deviation values.