Being a scientist, I believe most of my colleagues prefer to cite names exactly as the authors typeset them -- therefore, if there is a work by Karel Čapek, I would cite him as [Čapek, 1921] but I would not re-type a name like Ken-ichi Kawarabayashi back into the "original" spelling if the author himself writes it this way in a paper (e.g. [1]). And on top of that, sometimes (with say Japanese-Americans) you could re-type a name only to end up with an incorrect result!
This is not incompatible with the transcription+native (if any) approach: these researchers will most likely have papers or website with their name in their native language, while descendants of immigrants don’t.
Following your example, Kawarabayashi’s original name can be found in few seconds of googling. It is 河原林 健一. (https://www.jst.go.jp/pr/info/info915/shiryo2.html)
As an aside, I would love that more scientific papers use original names in their original scripts in the references section in addition to transcription, the way Knuth does it in tAoCP.