> [...] tedious to build [...]

It (or major parts of it at least) could be built by a fully automatic / robotized manufacturing plant. That isn't "tedious". Even for a one of its kind processor that would be much cheaper than the manual way.

> . It would be enormous, tedious to build (couldn't use modern lithography), and also consume tons of power.

Sounds to me like you're saying that Transistors are more efficient than vacuum tubes, both in terms of space and power consumption.

Many orders of magnitude so.

The last generations of vacuum tube processors were the size of large multi-story office buildings and had 50k tubes.

Compare to your smartphone, it would take billions of tubes to duplicate it.

Later research into thermionic devices did produce miniaturized versions, but nothing close to the scale of silicon transistors today.

If the same R&D might went into thermionics, we might just have devices of similar scales.

Yes. But at some point a quantitative difference becomes qualitativ.

If you're simply using efficient as a synonym for better, then every improvement is tautologically more efficient than what it immediately replaces. But solid state transistors were pursued specifically for their reduced power consumption while their scalability, which at the time was an afterthought, wound up making them revolutionary.

Tubes also burn out regularly (literally every day) when you have thousands of them.