Limits on capacity and number of drives is primarily going to be IO. IIRC, the USB bus is shared with the network, and they'll content for bandwidth. Not sure how else you expect to connect the disks, so yeah use USB adapters.

You absolutely won't be able to transcode on the fly. Not a chance. You can decode up to about 1080p @60fps but trying to transcode in realtime won't work, even if you compile for the VideoCore GPU and use hardware, it's just not enough.

I use a Raspberry Pi 3B as a Jellyfin server, and people are correct... it's not very suited to the task. Transcoding must be turned off in the settings as the the little Pi has no hardware decoding currently available, so I re-encode all my stuff to h264/aac before upload. Last I heard there were attempts to get some hardware decoding enabled, but I see very little progress. rpi-ffmpeg

I currently have 8 SSDs attached (which is probably too many), and I get about 12MBps throughput on USB 2.0. I have two powered USB hubs attached so that the SSDs don't undervolt the RPi, which is really easy to do. As people already mentioned, part of my speed issue is the ethernet being on the same internal hub as the USB ports. I recall that WiFi is not tied like ethernet so you may increase throughput by connecting to your network via 802.11. These rules change when you get an RPi4 though.

The media itself takes some time to scan in, and if you're booting from an SD card you'll want to mount all jellyfin directories on dedicated LVs to avoid tearing up your SD card. Also doesn't hurt to enable log2ram to reduce SD card wear, though you'll need to modify your system journal to flush more frequently so you don't fill it to 100% accidentally. log2ram

But yeah, after all that stuff it works more or less like I'd expect. The interface can take a second or two to load thumbnails, but I can watch 1080p content without buffering from inside the network... though content will always be exactly what you upload with no changes to resolution, encoding, or container. If you want different resolutions or encodings you have to upload multiple copies in the same folder (with specific naming) so they show up as options in the Jellyfin client. I even got some VLC installs to play the media by enabling DLNA.

I really would advise against a raspberry. Even a Pi 4 is too slow for transcoding and storage over usb is just too unreliable. I would go with a motherboard with integrated CPU. There is a really good one from Topton with a Celeron N5105 that supports hardware transcoding. I did a build with that recently and I’m really happy with it so far. Power consumption is around 35W with two 18TB sata drives. If you only use one and have it go to standby when not in use you could go even lower.

If you want something prebuilt there is a NAS from terramaster with the same cpu. It uses a an internal USB for OS storage that’s easily replaced. It’s called F2-423.