Yes, a bad cable has given me problems many times, with many systems (hard drives, USB, audio, etc.)! That's always a possibility. I try to keep extra cables just in case.
As for the amount of latency you'll notice: it will vary depending on how sensitive you are, how fast of a response you expect. I'm generally ok with up to approximately 12ms total latency. If I'm playing fast guitar parts though, sometimes my timing is affected by it a little bit and it's challenging to compensate.
Something else to consider: monitoring "through hardware" won't have the same latency as monitoring "through Reaper". If you can monitor from the audio device itself (instead of using monitoring on the track) during recording, that will be effectively "zero latency" (it's actually a few samples of latency, but nothing you'd notice). Reaper will record the part in sync with the other tracks, compensating for any latency during recording. If you are using guitar amp sims in Reaper (or are using MIDI synthesizers in Reaper), you'd need to monitor through Reaper. But you mentioned using a Kemper, so that should be something you can monitor in realtime through hardware. Check your settings of the RME card for its "input monitoring" (and/or "zero latency monitoring"). If you use that, your project can have 2 seconds of latency and it won't affect your performance during recording.
During recording, your system's latency (of the input) would still be determined by the latency of the Kemper and the Behringer, but the latency of the RME wouldn't be noticed.About buffer sizes: with my audio card, I initially chose 256 samples and 3 buffers (total = 768 samples). If I used 2 buffers of 256 samples (total = 512 samples), I would have problems. Later I realized I hadn't tried smaller block sizes but more blocks. I was able to make the system work correctly with a block size of 64 samples x5 blocks (320 samples), which is less than 2 buffers of 256 samples (total = 512 samples)! So the smaller size of the block wasn't as important as the number of blocks.
At 32 samples though, my system doesn't perform well with any number of blocks.
So remember that the total number of samples (block size in samples x number of blocks) isn't as important to keeping the system running correctly. There will be a minimum number of samples, of course. But maybe a smaller block size is ok, if you use more blocks. Also using larger blocks maybe is ok with fewer blocks.
Debian Stretch?!? What are you doin'?!? it's a commonplace i7 processor with integrated graphic!!! 
