A commons image is a work by itself, and there's not much derivation of
multiple files. OTOH data items are very tiny. So if made a graph from
eg. population numbers from different countries, if pieces of the data
were under a dozen different licenses, that could be problematic.
Yes, but in some jurisdictions the licenses are the only things allowing lawful use of the data (at least in sizeable portions). I'm concerned that we'd put some users, and maybe entire projects, at unnecessary legal risk if we ignore this.
If the graph in your example was based on 100 numbers from one database, and 100 from another database, then the licenses for both source databases could apply to the derived database. If the licenses weren't compatible, that would indeed be problematic in countries that recognise database rights. To me, this seems yet another reason to keep track of the licenses, so we don't unintentionally lead people into legal trouble.
On the other hand, combining data is not always problematic. If each population number in your example was from a separate database, extracting a single number from each database should not infringe on their database rights, and so their various licenses would not apply. (Of course, we'd still want to cite the various sources.) The resulting collection should be a fresh database that could then be hosted under any free license.