I’ve been running an n100 box as my main everything box for about 1.5 years. I capture metrics on it and can say the thing is nowhere near capacity. This box is running jellyfin, a dozen or so nfs mounts that are heavily utilized, a dozen or so lightly used samba mounts, grafana, prometheus, jenkins, and a handful of mysql instances. I maxed out the ram (32gb) from the start and it averages 8gb usage, and has never exceeded 10gb. Historically, the CPU usage averages 28% utilization. I think as long as the board has nvme storage you’ll not feel constrained by these little hosts for many years.
I’ve been using backblaze b2 (via s3fs-fuse container + bidirectional mount propagation to a host path) and a little bit of google drive (via rclone mount + the same mounting business) within kubernetes. I only use this for tubearchivist which I consider to be disposable. No way I’m using these “devices” for anything I really care about. I haven’t tried gauging the performance of either of these, but I can say, anecdotally, that both are fine for tubearchivist to write to in a reasonable amount of time (the bottleneck is yt-dlp ingesting from youtube) and playback seems to be on par with local storage with the embedded tubearchivist player and jellyfin. I’ve had no issues with this, been using it about a year now, and overall I feel it’s a decent solution if you need a lot of cheap-ish storage that you are okay with not trusting.