Well, you're also forgetting Amdahl's law:
1) Rule of thumb: 90% of the time is spent in 10% of the code.
2) If you speed up the least-used 90% of the code by an infinite amount, your program becomes 10% faster.
3) If you speed up the most-used 10% of the code by an infinite amout, your program becomes 90% faster.
In other words, for 1/9th the effort, you get 9 times the speed up. If your program takes 10 seconds to run now, the first option might require 9 hours to get the run time down to 9 seconds, while the second option would only require 1 hour, but get the run time down to 1 second. Also note that this assumes that you're able to speed up selected portions of the program by an infinite amount; when you look at more practical scenarios, the comparison becomes even more stark.
These numbers are ideal, but the lesson remains: Programmer time is a limited resource; you can't economically afford or physically obtain unlimited programmer-hours. Best spend that time on what you can improve most effectively.
If the compare the cost of writing the very small amount of code which ran the Moon computers to the cost of the word processor you run on that 1 GHz machine, there's a clear win. Of course, the word processor has developed over the years and the 1 GHz machine didn't appear overnight, but in today's world, the trade-offs are not only inevitable, but make a lot of sense.
Coding something with the complexity of modern programs completely in assembly language would lead to more bugs, cost more, take longer, be harder to fix, and be less portable to new architectures (Privateer on Windows XP, anyone?). You just don't win when you waste human resources like that.
And anyway, my word processor doesn't require a 1 GHz machine to run. I used to run my word processor happily on a 100 MHz machine when everyone was buying 500 MHz machines. Microsoft will bloat their programs, but even they don't really use all the computational power we have today. We get 2 GHz machines not because we need them, but because they cost about as much these days as the 1 GHz machines.
Put this another way: would you rather have that time spent on including a decent storyline and the pretty new models, or would you have them spend their time searching all the code for useless text strings which only added a few bytes and microseconds of download time to the compressed version of Secret Ops?