There's a few things going on here:
Saying that there are 9000 unconfirmed transactions is not a particularly good fit for how node memory pools work. Not all pools contain the same transactions, not all nodes have the same rules for what an acceptable transaction is. Operators are free to modify parameters that alter this on a case by case basis, and memory pool contents will not be retained across a restart.
Increasing the block size is not a solution to spam transactions on the network. If you keep making the limit higher there is no reason in the world people won't produce transactions to fill the space. Fee prioritization forces spam to have larger fees than other users which limits how much they can create for a particular budget, the price for a user to get above the spam is in the order of a couple of US cents.
Miners don't have to include any transactions at all if they don't want to. Some are paid out of band to include specific transactions in favor of others, some choose to only include fees above a threshold, some refuse to include free transactions at all, some choose to ignore transactions with specific properties. Not including intentionally spammy transactions that bloat the UTXO database is a desirable property, where it would be harmful to all network participants if miners mindlessly included all that have been broadcast.
"Stress test" is a bit of a cover term, the event you are referring to is a denial of service attack. If there was knowledge to be gained it would be happening in a testing environment, not being announced with the intention to disrupt use of the network.