I agree if you mean that the damage from an irrational bias is higher when the stakes are higher, but disagree if you mean that rational marginal certainty levels needed for conviction would be higher for severe crimes. The risks from letting a thief go free (more thefts) seem lower than the risks from letting a murderer go free (more murders) even compared to the damage done to a potential convicted innocent (assuming no death penalty, and also assuming higher conviction rates would actually result in fewer of the real culprits going free, which often does not seem to be the case).