For me, I have avoided all HA/replication (with the exception of backups). In my 20+ year career as a professional system admin/engineer/architect in increasing levels of responsibility/authority I have only seen HA/replication cause more issues than it's worth.
To be clear, I am referring to things with state (databases). Starless (application frontends) and clustered "semi stateful" (think memcached/redis) is quite welcome/acceptable.
Database replication with appropriate monitoring/resiliency/planning could be useful. It can also go sideways in nasty ways.
I have found Cloudron backup/restore to be quite fast (presuming your Cloudron instance is network close enough to your backup target). The upcoming backup changes are quite welcome and address all of my concerns.
Add a CDN in the front and enjoy stateless/horizontal scaling.
As I understand it, swapping out your data store to a cluster (and keeping the docker/readonly/app bits in Cloudron) should be easy? Just need to update the DB_ related environment variables?
I (and my company) are very heavy users of Cloudron. Being on a single box hasn't been an issue for us. When we need to scale (we expect to have some massive read heavy apps using some complex GIS stuff) we will use a CDN. We are spinning up a large k8s cluster to run a number of high compute workloads. All of the command/control will be via Cloudron hosted apps (BOINC/SLURM).
Cloudron is for "bootstrap/core" "pets" (but using a kind of "cattle" architecture/model)(this combination is very powerful) , k8s is for your scale out cattle. (k3s/rancher/longhorn) makes k8s deployment quite easy).