Herbie writes

 > don't be that pessimistic...

After crashing FB and crashing myself, my troubles
with UNIVERSALPROC & UNIVERSALFN evaporated.

The internal variables of the procedure seem to be stack based
locals -- just as for FNs.  

Just some marginal notes which I hope someone will correct if
necessary. The PROC type for the proceedure pointer, and the use
of PROC "label" rather than LINE "label" to set it up, seem
to be a must for correct compilation even though
you will find that they give one and the same address value.
The commands CLEAR and MODE as used for FN's are not available
even tho the compiler accepts them in some forms; they just 
cause crashes.  The first 4 bytes of the local variables 
seem to be always cleared. But the later bytes never are, or rarely.
That started some of my delusions.

As for Carbon support, would Andy Gariepy be the first man to
know?

Other than that, my main unanswered question here is

 > Another (old) question:-  Has anyone been able to set
 > up a UNIVERSALPROC or any such device that is able to
 > exploit binary PPC code **coming from C** and stored
 > in a resource?

Cheers

Laurent S.