The benchmark is of limited use to determine requirements since we do not know how to translate these numbers into real-life features. But at least you can use the numbers to get a rough idea of how powerful the various CPUs are.
In addition to these numbers, I should state that the i3 CPUs are marginally less efficient (more heat per work and therefore more noise) in this test than either the i5 or the Pentiums of the same generation.
however imo i think i will need the i3. I've read that some people who have a similar setup like I want (Plex running on a server and streaming to a roku box) run i3's or i5's.
Don't mind what people use. People are generally clueless.
this rig will be in a bedroom and it probably will be transcoding often when someone nearby sleeps, so i want it to be as silent/stealthy as possible without breaking the bank or sacrificing a lot on performance.
There are no miracles. If you aren't willing to pay for a passive build (that would require an expensive case such as the one SPCR just reviewed but 500$ should be sufficient if you spend carefully), consider changing the server's location (you don't need room on the floor: you could hang it on a wall, put it on top of a cabinet or something) or ditching the real-time transcoding requirement: can't you store your files in the encoding needed by your devices to begin with?
Note that passive builds are generally not supported by motherboard manufacturers.
when you say a DC powered board, do you mean like a board powered by a PicoPSU?
DC-powered boards have something like a customized and efficient picoPSU on board, except of course you get a configuration supported by a major manufacturer.
You simply connect a compatible AC/DC adapter to the motherboard and it powers the whole PC (no PSU).
If you want an i3 or an equivalent Pentium/Celeron, Intel has Sandy and Ivy DC-powered boards. They're kind of pricey but the price includes the picoPSU-equivalent and the efficency is useful for passive builds.