This has been a valid approach in the past, yes: region around the noisiest/jaggiest part of the image in final resolution, then region render until it looks good enough to get your pass count.
However when using denoisers of all kind (including Coronas internal) I don't know how region render affects the result.
Why not just Dr this image?
Good Luck