Exactly.. Something like pi-hole which is self-containing (so doesn't need network shares and fiddling around with bind mounts) or jellyfin (which can be installed with apt install under Debian) where I want to use the host hardware (for jellyfin iGPU) but also share it with other lxcs/the host (I could passtrough the iGPU to a VM but then I wouldn't have a console anymore for trouble shooting if the WebInterface gets broken) are great usecases for LXC.
My monitoring software (icinga2 + graphite at the moment, I'm considering switching to Prometheus+grafana though) is also installed in a lxc, since everything was installed with adding the icinga+graphite repos followed by an apt install $package.
Another usecase is if your install of ProxmoxVE is on a vserver or inside a vm without support for nested virtualization so you can still have several environments for testing/playground/seperating applications. Maybe you want to have several instances of the same application (like a webserver for wordpress, one instance for testing/developing, another one for production).
I remember that one of the Proxmox developers mentioned at some point, that one of their customers is running multiple hundreds of lxcs on the same host, which all contain the same software stack to host the website of the schools in their region:
Please note that the quote is from 2018, so right now even higher numbers might be possible (I don't know whether the mentioned technical limit of maximal 3000 vm or lxcs in one cluster are still relevant) Interesting enough in the same topic fellow community member sprit (who also developed the lvm snapshot support introduced in PVE 9.0) mentioned that at that time he run 200 VMs on a 2x12 core node (he didn't mention the available memory tohough). Anyhow I think it's save to assume that even on a Mini-PC-"Homeserver" the available resources should be enough, given you plan accordingly how to utilice them
Another example:
On a technical conference I met a guy who is using lxcs as generic tool to provide his customers with managed webservers. I asked him why he don't use VMs (since they allow downtime-free migrations and provide better isolation). He pointed out that this allows him to fit more customers on one host, and lxc downtimes are small to beginn with. For any customer who can't tolerate even some seconds of downtime in case of an migration he would use a loadbalancer like caddy or haproxy. What managed webhosting usecases all have in common, that they are not using docker containers but are basically multiple instances of the same basic system (a Linux distribution with an installed webserver, scripting languages (php5/python/ruby etc) and a sql database). So the lxc containers are to isolate the customers from each other but still allow dynamical allocation of the resources of the host system. Since the system is nearly identical mentioned webhosting admin also automated the maintenance with ansible to avoid the overhead of manual administration of every container.
Obviouvsly you could also use this approach for VMs, it really depends which tradeoffs in terms of resource usage/security considerations/maintenance etc you are willing to take. German webhoster uberspace run their plattform on CentOS VMs in a ProxmoxVE cluster managed with ansible. They write a lot about their setup in
https://blog.uberspace.de/ (if you understand German). Basically every vm has the same setup of CentOS (they plan to migrate to Arch but they are not there yet) with a webserver, several scripting languages and mariadb/postgres. The webserver is configured for shared webhosting, so on each VM they have a couple hundred users with their websites, several of these vms are on one node and all nodes together build their ProxmoxVE cluster. The same setup could propably also be done with LXCs to fit even more customers , but the VMs provide stricter isolation which is a good thing from a security point of view.
Any of the mentioned corporate scenarios (with lxcs or vms) are however not typical for homelabs imho
