Do you really think it would take 30 years? The NAND-To-Tetris kids do most of it in a semester, and they're starting at the level of individual gates.
I mean, Chuck Moore has been doing it for about 30 years, but he had a working top-to-bottom system after less than five years.
Here's what I think it would look like:
Week 0: build a working interactive Forth system that can compile itself, in machine code. This is assuming you already have hardware and a way to boot it into some code you wrote. If not, that could be a matter of minutes or of months.
Weeks 1-3: build an interactive high-level language interpreter on top of it. Say, JS.
Week 4: enough of TCP/IP to do HTTP GET.
Week 5: write a bitmap font editor and a minimal filesystem.
Weeks 6-8: parse HTML (more or less), lay it out with a simplified box model, and draw it to pixels on a canvas, using a pixel font.
Week 9: cookies, HTTP POST, XMLHttpRequest.
Week 10: a more elaborate filesystem, maybe based on Git.
Weeks 11-12: some more HTML5 features, maybe including WebGL.
Weeks 13-14: TLS.
This is sort of cheating because you're taking advantage of our previous 70 years of experience to figure out what's worth building, and you're likely to end up spending a fair bit of time going back and retrofitting stuff onto previous iterations — JIT compilation, say, and multitasking — but it still seems like rebuilding the personal computing environment ought to be more like a one-semester project than a 30-year project.
You could argue that you get a huge speedup by not writing stuff in assembly, and I agree, but I think that's only a constant factor. You could easily write assembly code that maps directly onto JS semantics:
;;; invoke getContext("2d") on the canvas object
.data
sgetContext: .asciz "getContext"
s2d: .asciz "2d"
.text
mov $sgetContext, %eax
call GetProp
mov $s2d, %edx
call InvokeMethod
That's more code to write and read, but only by a linear factor. So it might take you two semesters, or four, but not 60.