Good point. But I don't have any better solution to offer for now...
The way I build this structure many years ago was to give the option share the channel file across folders within one subject, or across all subjects. But the channel file includes the SSP (which are typically different for each recording session), and goes with the noise covariance matrix (which cannot be shared across subjects) and the head model. So in the end, the channel file + head model sharing is not really possible...
To copy the headmodel from a script, you can find some inspiration in the script attached to the group analysis tutorial: