Of course not. But here we are looking at a 100x slowdown.

Yes it can. On some controllers. There's obviously no improvement when
writing a file, but when reading, the closest head to the requested track
gets the call.

If one head is on track 10 the other head on track 90 and the requested data
is on track 90, then head 2 will be assigned the read, eliminating the
access time necessary to move head 1.

I might do, but will the PC still boot?

I have found that the format of a HD which was in a mirror raid array
is not the same as the format of a HD which was used alone.  Why this
is, I haven't got a clue since it isn't necessary AFAICS. The PC does
not even boot from a HD which was in an array and is then moved to the
standalone config on the *same* controller. So if I change the
controller to another brand, I might lose all the data...

SIIG just use the Silicon Image SIL3132 Softraid 5 chipset which lots
of firms use. The SIL31xx chipset is all over the various motherboards
from Gigabyte, ASUS, etc. In this case I am using a standalone RAID
controller (2x SATA 300, PCI-Ex bus) and even I bought a spare one, to
avoid this kind of dependency. My motherboard happens to have a SATA
RAID controller already and probably with the same chipset, but I
didn't use it for this reason.

On another machine (critical one, used for accounting) I used an
Adaptec SATA RAID (150) controller and that one works fine, but in
this case I wanted a 300 one and Adaptec don't seem to make those.