The architecture can work if enough smart people are put to work on it. That's how Apple managed to turn macOS into a mobile operating system.

I think UBPorts and Sailfish prove that Linux for phones is practical if you're willing to rely on Linux applications that stick to mobile friendly APIs.

You need to configure and compile your Linux kernel for aggressive power saving, of course. Seeing how Linux currently struggles to effectively do power management on laptops without S3 sleep, there's plenty of work to be done if you want to use it with a phone.

It's not just about app developers either, Qualcomm's modifications to the Linux kernel are public thanks to GPL but most phone kernel modifications haven't made it into the upstream kernel so far. Projects like postmarketOS are trying to make things better but it's not easy to port practical code that works into code that's acceptable for the maintainers of the broader Linux project.

SailfishOS also came (at least back in the day of the first Jolla Phone and Tablet) with an excellent terminal app and built-in sshd that made it work great with pretty much every Linux command-line and TUI application (only exception was of course those with hardcoded minimum screen size support). Termux for Android is maybe half that good, not as well integrated, but still good enough that I use it every day, much more than I use other apps other than the browser.

But I mean, why not take the 100s of thousands of man-hours that went into making Android this very Linux-Kernel based mobile OS and build on top?

Will you be happy with xeyes and a terminal? Like, even a technically superior solution is completely useless without an ecosystem to make use of it, and desktop GTK/qt apps won't work nicely on mobile without actual porting. Let alone a technically significantly inferior one that is a misfit in this shape for mobile hardware.