Nothing is impressive until there’s 3rd party testing. Wouldn’t be the first time a company made big claims only for the arch to shit the bed in other workloads.

the power constraints would not make it possible

This is pretty much your answer right there. The Series S is a high throughput design, packed with shaders and CPU cores with relatively little cache per core but high bandwidth high power GDDR plus optimized software. Van Gogh is a low power design where the shader count is pretty much limited by the power envelope. A large part of the chip isn’t even shaders or CPU cors but other fixed function special purpose stuff. AMD could probably double Van Gogh’s throuput with +10% die space but that’s pointless because the power budget isn’t there.

That’s 210GB filled for a 1000GB drive

No, that’s 630GB of TLC space filled. Writing in SLC mode requires 3x the space. The dynamic SLC caching stops at some point because the controller still needs space left to rewrite 630GB down to 210GB during idle time plus a safety margin so the user won’t run into a situation where the drive has to “freeze” to catch up with the work. There is always some minimal amount of SLC cache available through overprovisioning, but that’s typically only a few GB.

Water cooling is a convoluted solution for outdated 90’s case designs. Why would you need water in a modern airflow case which can easily exhaust 1000W+ of heat with fans running at 500rpm or so?

No gray area, at base clocks the 4090 exceeds the limit by 10% already.