@pspacek It's happen periodically that a site is live and reachable, but his NS is down or unreachable (for example, when network connectivity between target NS and my knot-resolver is broken). In this case, site will not opened because DNS resolution via recursion is failed. Forwarding of failed request to 18.104.22.168 can be a good fallback because/if: (a) 22.214.171.124 returns cached info, (b) network connectivity between 126.96.36.199 and target NS is ok.
@ilyaevseev right, sounds reasonable. I remember a few occasions when such a fallback would prove handy. Some remote nameservers blocking traffic from chosen networks for no obvious reason. An externally placed resolver could help in such a situation
@vcunat serve_staled cannot help us on the first request failed.
@pspacek forwarding itself is faster, but public DNS forwarders (Google, Yandex, Level3, ...) are relatively slow. And their answers may contain old info, so we cannot use forwarders as a primary source.