Can you help me figuring out what I am doing wrong?

Longhorn uses iscaci NFS under the hood, as I understand it.
=> uptime-kuma contains a database
=> you are running a database on a network share
=> possibly the added latency of reads/writes is killing the database performance and not #3515 or

Note that running on a NFS-Style system has soundness bugs with SQLite databases due to faulty file locking, which may lead to corrupted databases.
Please run uptime-kuma on a local volume instead.
See https://github.com/louislam/uptime-kuma/wiki/%F0%9F%94%A7-How-to-Install#-docker and https://www.sqlite.org/howtocorrupt.html#_filesystems_with_broken_or_missing_lock_implementations

I am running Uptime Kuma on a Kubernetes cluster with 3 Servers

HA will not work with uptime-kuma. Please don't run multiple instances of the same docker container as this may corrupt the database.

V2 includes a version to connect to external databases (or continue with the embedded mariadb/sqlite)
See #4500

• better performance through external MariaDB, storing heartbeats in an aggregated form, full server-side pagination for important events
  => even though benchmarking is still out, we are confident that this pushes the prominent "limits" (highly hardware-dependent, not a limit imposed by us) ~500 Monitor or ~1.5GB DB-size
  A large part of the focus in this release is on performance. We don't know if we are optimising for the right thing, see Can/should we add privacy-respecting usage metrics? #4456 for a discussion on this topic.

In the meantime, choose a lower retention to mask this issue.

@CommanderStorm Thank you for your fast and insightful comments! It is really helpful.