Did you read binmode? (Granted, it is rather inaccurate and shows that the author doesn't understand the point -- click on one of the links to more modern versions of the document for much better text.) It says "Files that are not in binary mode have CR LF sequences translated to LF on input", which is accurate for Win32 systems. Checking for such takes some time. Actually having to fix that requires that all of the text in the buffer after any CRs needs to be "moved up", which takes even more time. Checking the source for the standard Microsoft C run-time library, I see that the standard:

    char *in, *out;
    in= out= buffer;
    ...
        *in++ = *out++
[download]

method is used to avoid multiple "move" operations, which means that the cost of moving is incurred even if no CRs are found [ and also makes for much simpler code that is easier to "get right" than if we tried to switch to strchr() and memmove() to allow assembly-language constructs to search the string and to move the bytes (: ].

        - tye (or ldad $54796500 if you're in a hurry)