Batteries. Running a full Ubuntu OS on a smartphone battery would result in the battery draining quickly. So instead you'd need some sort of dual boot phone OS and desktop OS system that shares data... it's far from simple.
One OS should suffice. Driving a desktop monitor and powering desktop app features while still on battery would certainly drain battery life, so don't do that. Plug it in. The OS and apps will morph to the I/O and power class it has access to (we already have power-stepping processors... it can't be that far-fetched).
Also, from this whole argument it simply doesn't follow that one OS can't scale from IoT to mega-servers. It's just code. Not all of it has to be running all the time. Not all services have to be proffered at the same time. We can do that. It may not look like "Ubuntu desktop OS on a phone", but maybe it shouldn't. That's probably the wrong direction.
Yes. Turns out the cost of context switches quickly overran the benefits of this approach, but things are slowly getting better. L4 is one approach. The library-OS approach (where apps run directly on virtualized hardware with a minimal OS) are another.
This must be some strange definition of OS. Phones already run on a Linux kernel - and are already capable of running a GNU/Ubuntu yserland in parallel with an Android userland. Clearly the Android userland works quite well for mobile use, and the desktop userland could work quite well for docked use. It seems odd to claim that the two modes of userland couldn't be more tightly coupled (allowing a single app to expose multiple UIs, like a browswr exposes both a command line interface (firefox www.google.com) and a GUI?
Batteries. Running a full Ubuntu OS on a smartphone battery would result in the battery draining quickly. So instead you'd need some sort of dual boot phone OS and desktop OS system that shares data... it's far from simple.