For loops, even in C# are used in performance sensitive contexts. I believe things have changed in versions of .NET and if you're using things like Mono that also have its own versions.
Foreach generally (or used to) generate more garbage/allocations. Foreach against some types in some versions of C# (ex: structs) produces no allocations. There are also some ugly mutation related things you could do using for instead of foreach, but I'd call that bad code.
The same things are true on many platforms - foreach is generally preferable for non-peformance sensitive code. In some languages and libraries, foreach constructs will do things similar to auto pointers or reference counted pointers, which makes them safe enough, but also slower. The point is that foreach can take up more memory, cause garbage, or cause allocations/free on some platforms and languages which is not always desirable. For is therefore a better choice if you don't want this behavior. Mostly though, using for instead of foreach is micro-optimizing in these cases. Where I'd usually do it is somewhere critical in my code, like something called a lot per frame game loop for example. If I caught someone using a foreach that generated lots of allocations and is called a lot of times in these contexts, I'd probably kill them (and have).
So in short, I guess I'd agree that "many" people would do it, but it's context dependent which include performance goals, target platform, language limitations, and more. Mostly, it's better to just write the code that is safe and works, and go back and fix any bad decisions like this. If it's obvious though, I don't have a problem with the optimization from beginning.
Having made my fair share (if not more) of terrible (and terribly embarrassing) errors in C using the for-loop, I love foreach and its brethren in other programming languages. I only write classical C-style for loops very, very rarely these days.
