Who cares about
Turing completeness? You're running on a machine with a finite
number of states; so a DFA language is all you can really
implement in real life. Turing complete languages can't be
implemented; just approximated.
I could attack this statement from the theocretical side.
Computer hardware is developping in a very fast pace.
Let's imagine you have a system where
the state of the program can be saved and reloaded on a
machine with more memory,
or one that runs on a transparent
cluster of machines where new nodes can be added dynamically
and old nodes discarded.
Such a system could run forever, if you keep updating it with
new hardware.
While my lifetime or the lifetime of the computer I'm typing
on right now is certainly finite, such an advanced system
could live forever, if the civilisation doesn't end in
some catastrophe.
(In fact, Paul Davies argues in his book
The Last Three Minutes that a super-entity
could do an infinite number of calculations,
and discusses this in the two cases of whether the Universe
will expand forever and end in a Big Chill, or
gravitation will win and the Universe will end in a
Big Crunch.)
However, even if a DFA is all you can implement in the
real-world
(there are some steps between the DFA and a Turing
machine in power of course),
I am a mathematician, and I'd like to at least imagine that
I have a Turing-complete machine.
I use it even if it's only a theocretical construction.
I am not an engineer, who uses a ruler to draw straight line,
and uses large plastic templates to be able to draw any
kind of ellipse.
I just draw a bumpy line or a potato by hand, and say it's
a straight line (moreover,
an infinitely extending straight line that has no width or
thickness) or an ellipse.
This is a very pleasant, almost god-like state,
because I can create objects engineers can't even dream of
just by the power of my mind.
Naturally, this is not the real reason I wouldn't rely on
foreach loops.
The reason is simply that they're often difficult to use.
It's not easy to convert while loops to foreach loops.
It's even more difficult to do it if you care about efficency
(because I don't just have any DFA, but a DFA with a
limited number of states and limited speed of execution).
Foreach loops require to use side-effects even if I
don't want to, and you could easily avoid it with do loops.
Foreach loops come in different versions
in each programming language.
But the deciding reason is that C doesn't have a
foreach loops, so if I decided on foreach, I couldn't use
C anymore.
Update: "it's silly to stop with a Turing-machine" -- you're
right with this part.
"Why not use a bigger field(?) than the integers" -- indeed, why not:
Symbolic calculations with operator overload, Re^3: Illegal Modulus zero.