There is no specification of stack usage in the datasheets so I set SP to FEh (just to accomodate the return address for LCALL 0FFF0h) ... And it fails to program, crashes etc. (OK I did not try this but I suspect it will).
It will

What did I wrong?
Create a case just to state a point that otherwise is meaningless.

As a peace offering (to get this thread ended):
I can see all the points/uses mentioned, just can not see the original IAP code creators having them in the specs. Most of us have been asked to create something that could do 'x' just to be asked, after completion, "why can it not do 'y'.

So, the solution I'll recommend (as I have done before) is to rely on XRAM for whatever need be carried over and just do a reset after using IAP. This may not be "the best thing since sliced bread", but it works.

Erik