Speaking in practical terms, the point is not actually to make it completely impossible for anyone to access data they "shouldn't". The point is to raise the cost of access - in effort - beyond the point where it's economically viable to go after any but the biggest targets. In short: So a dedicated group could compromise you with a computing cluster... So what?
If you don't find that reasoning palatable, consider this:
Try and think of an estimate, in dollars for services rendered, to do the following:
1. Have someone steal your phone
2. Disassemble the phone and read the contents of the flash RAM and the secure enclave out (in the latter case by dropping it into an acid bath and manually reading the status of the bits out of the traces - yes it can be done) (remember, the password only permutes another, much longer key in the enclave)
3. Pass this info to a good-sized computing cluster
4. Dig actionable intel out with some good forensic software
Now compare that dollar cost to what you might pay some local thug to:
1. Hit you with a brick until you give out the password, or in the case of touch-ID, wrestle your finger onto your own device.
If the cost of scenario A is higher than the cost of scenario B, then problem = "solved".
Unfortunately for you, even if you come up with some epic convoluted method to render BOTH scenarios totally unfruitful, as long as scenario B works _some_ of the time they will try it _anyway_. And you will probably end up dead.