Why does rndc reconfig sometimes resolve recursive server problems with some domains?
There are many poorly-configured DNS servers that provide records that break the DNS protocols. Depending on how poorly these servers comply, name resolution may work most of the time. Factors influencing this include the relative TTLs of parent and in-zone NS and glue records and what has been queried for the zone already (thus what is in cache when the next query is received). Improper use of CNAME records can also cause unpredictability (for example, having an NS record that points to a CNAME instead of a hostname).
While the onus is on the zone administrator to fix their DNS servers, sometimes it's necessary for operators of recursive servers to clear a problem for their clients - even though this is not addressing the root cause. The problem will most likely recur again.
Typically, they will use rndc flush or rndc flushname to clear the cache of all related names.
Sometimes an rndc reconfig is more effective. As well as re-reading the configuration file and re-initializing various internal settings from it, rndc reconfig also clears the ADB section of cache.