Before wheels, I'm pretty sure people used tree trunks to drag heavy objects from A to B. And I'm pretty sure the wheel's inventor hasn't envisioned the average consumer car we have today either.
Linus's point is that design happens at a micro-level, but at a macroscopic level the direction of a successful project should be made by trial and error with a feedback cycle.
To put things in perspective ... Linux is not designed for servers, it is not designed for desktops, it is not designed for real-time systems, it is not designed for mobile phones. When people wanted to use Linux for such stuff, they took it and modified it to their will, and the parts that slowed them down were eventually removed / refactored.
That's actually very true if you look at the history of many successful projects. And that's evolution more than design. "Release early and often" is exactly about this.
Well, no it isn't too good. But it does basically fit: It is a small feature of larger systems, as you describe, and it has been done better by artificial design.
design at the micro-level = hypothesis formulation direction of successful project at macroscopic level = repeated testing and reformulation of hypothesis
in other words, project success is science well applied
Evolution got there repeatedly. Legs just work better cross-country, and there didn't use to be much roads around.
The reason, of course, is that our guts are adapted to pre-processed food, and so are shorter. Therefore unless you get some cooked nutrition, it is very hard to get enough food. It is possible, some people do it, but it is hard.
Saying "wheel!" is crude and invites a simple counter. But the point is there is a simple, important thing where artifice has a substantial lead.
And without well-chosen application of design, the large-scale evolution might be practically excluded from certain areas.
(. . . that was really a side issue. It was the procedural similarity of design and evolution -- they are not so much opposed, as differently balanced.)
Linux as a whole was not designed -- that is easy to accept, and also that it is a strength.
However, design is not so entirely opposed to evolution as might be inferred. Design seems essentially to have some iterative character too. It always entails a some series of ideas, rejections, and then further adjusted ideas. That is deliberately controlled, but it still has variation and selection -- so the possibility of blending between design and evolution is very reasonable.
And given the wheel example, artificial design has something to offer too.
(I think Torvalds' view is consonant with this. -- I mean, is he not overall offering some combination of evolution and design?)