So, let's look at your claim by way of example...

Linus gave a talk here at MIT a few years ago, shortly after the Alpha
port was initially released.  I asked him how long it took.  The
answer was 7 months.  Now, consider that this was to an EISA-based
machine (the Jensen, I think), for which almost all of the devices
were already `supported' by Linux already.

By contrast, our Alpha port was up and running fairly reliably in 3
months.  It was initially done on TurboChannels machines (which,
AFAIK, you folks *still* don't run on), which required writing a
considerable amount of new device support.

`Which approach took less time?'

Also, I can plainly state that in my experience it would have taken
substantially longer to change all the kernel structures to get a port
running.  For someone with knowledge of the particular CPU's assembler
code, the entire sum of all the machine-dependent user-level code can
be written (although perhaps not optimized well) in a day.  It's just
not that hard to do.

In short, I think you're puffing hot air.