a margin of 12dB is good enough, because the donor signal is indeed considered as an interference. The delay between both signals is too long to be considered as multipath. (of course, i said that based on my theoretical knowledge. perhaps there are some repeaters that don’t face this problem because they repeat the signal very quickly).
there is something else you should consider (but i’m sure you did already) : the amount of feedback that leaks from one repeater antenna (ex: donor side) to the other repater antenna (ex: indoor coverage).
You should measure an isolation that’s higher than the repeater gain + 12dB.