As I'm writing this, I've just migrated the final PVC to piraeus. I've tested things with smaller volumes over the past week and have now migrated my forgejo storage volume, which was the last remaining jiva PVC. It took a bit of messing around with the helm chart to get piraeus to a state in which I like it but the end result is pretty convincing. It is light on resources, backup works well and workloads can continue to write to the pvc even if the replica goes down.
It turns out, this was also a huge performance increase! Forgejo is much more responsive and git operations are near instant. Interestingly, the metrics gathered by the ingress show a slight increase in the lower bound for request latency but the request with higher latency are drastically reduced as well.
I'll try to finalize scheduled backups today and then finally deploy a matrix server again. Another open topic is the migration to HAprox as ingress controller & setting up iocane to reduce load on the forgejo instance but that's topic for another time :)