(I seemed to reply only to Arthur, so I resend.  Sorry if you get duplicates)

I've seen Per's post more of a interesting challenge than a proposal...
Thought of trying it during past weekend but couldn't get around other chores.

To comment to it, I tend to see a quasiquoted form (with unquotes) as
a piece of code rather than data, and having cycles there is more like
having cycles in the code---that is, implementation may diverge.
But that's just a feeling; it may be useful if we allow cycles in "static"
parts of the template.

Btw, if we expect quasiquote expander to recognize shared structure,
what should the following code yield?  Specifically, do we want to mandate
the first and second element of the resulting list be eq? or not?