It still doesn't answer the question however of "Does the CBO use the stats seen in all_tab_modifications for query plans?"
I have never seen such a behaviour (but I haven't tested it yet)It might be of any sense to improve the affinity for sampling data about a table if the count of modifications is too high, but - never seen something like this.
maybe you can create a testcase (best with a fresh started instance or at least flushed dictionary cache) and event 10046 trace: in this case you will see a lot of queries from sys, where all the information the kernel requires about an object are selected (and loaded into dictionary cache). If the base-tables for *_tab_modifications are part of these queries, they might be of any interrest for any kind of the kernel (not automatically said the CBO!).
Otherwise I would be quite sure, at query time noone cares. hth Martin