David,
Very interesting question! Let me think aloud for a while.
A reply or an update can remove an entry from the show ip eigrp topology all-links if and only if it is a poisoned information, i.e. the metric is infinite (usually, the DELAY component is set to maximal value). Otherwise, the update or reply simply conveys an information about a network as known by a neighbor and reachable by it, and there is no reason to throw away such information or to remove other entries from the EIGRP topology table as a result. At most, this information will not pass the feasibility condition check but it still can be recorded in the topology table.
A query basically conveys information about the current (increased) distance of its sender in the moment of sending that query, along with the request to reevaluate our own routing table and send a response. If a query is received, our routing table may stay unchanged (if the query did not cause us to lose the current successor), or it may change in the process of the diffusing computation if the query came from the current successor and violated the feasible condition check. In any case, the network stored in the EIGRP topology table for which we received a query must be retained in the topology table until all our own queries have been replied to because the state of the network (active, waiting for replies, etc.) is recorded along with this route in the topology table. In addition, the current (perhaps increased) feasible distance of this route may be used after all replies have been received to perform an additional feasibility condition check, so under no circumstances can this entry be removed because some vital information needed for properly terminating the diffusing computation could be lost. After all, in SIA states, it is possible to see the networks in the topology table marked with 'A' (active) for extended periods of time. They would not be there at all if the arrival of a query removed them from the topology table at all.
Best regards,
Peter
