I think the catch here is that under STUA (short term use address) there is a strict incentive, you can reduce the transaction fee for these txns. This also fits with the general model that you pay the miners for security. My belief is that when there is a savings benefit to be had large players will prefer it at a minimum, and users will desire it.
Your analysis of saving is inaccurate, it comes to be at or greater than 20 bytes because there is typically 2 UTXOs or more. I get that this is still marginal, but when the margins are tight this is a pretty decent gain.
The security decrease is actually less extreme than it seems. This is for multiple reasons:
1) you can select LEN_PARAM when you make the tx to be secure at that time Adding a byte or two gets much more security while still keeping it lean.
2) For a small transaction, the hash power is less rewarding than just working on the blockchain or doing something else
3) These addresses are only for use for short term, not perm storage. As such, if you model the threat it isn't great (I'm using this address for one day, someone grinds it in that time).
4) Because it is a UTXO saving, it reduces memory bloat.t
It would be interesting to get a more exact analysis on the time needed to run a brute force as it involves computing a valid keypair and hashing for each attempt.