currently, 2 cvmfs-atlas containers won't share their cache even if they run the same exact command atlas read-gen-event.py.
one way to fix this is to put a squid container in front of the cvmfs-xyz containers.
another way is to VOLUME-mount CVMFS_CACHE_BASE=/var/scratch/cvmfs in all cvmfs-xyz images and provide a data-only container exporting that VOLUME.
AFAICT, all the possible solutions will require a 2-container (or pod-like) setup.