If the client manages to get a successful response then it is cached normally - whatever the TTL value attached to the record defines. In case of PendingDNS all

If the client manages to get a successful response then it is cached normally - whatever the TTL value attached to the record defines. In case of PendingDNS all TTL values are 5min by default. So the problem is with unsuccessful responses - there is no TTL attached, so the client uses whatever it finds suitable - maybe uses values from SOA records, maybe has its own default, who knows - in any case, it is quite long.