In GNU Emacs 23.1, there is a problem in decode_eol of coding.c.

In buffer of which enable-multibyte-characters is nil,
the function `decode-coding-region' should delete the character '^M'
(code: 0x0d) at the end of line,
in case that the function parameter `coding-system' is *-dos (and
the variable `inhibit-eol-conversion' is nil).
But, in practice, the function doesn't delete all of the character '^M'.

You can watch the problem with the following.

========================================================================
(progn
  (pop-to-buffer (generate-new-buffer-name "*scratch*"))
  (set-buffer-multibyte nil)
  (insert (encode-coding-string "あ" 'euc-jp) "\xd" "\n")
  (read-char "(press any key)")
  (decode-coding-region (point-min) (point-max) 'euc-jp-dos)
  (read-char "doesn't delete all of the character '^M'. (press any key)")
  (set-buffer-multibyte t))
========================================================================

I made a patch to fix the problem. Please check it.