FWIW, another experiment that produces unexpected results:

The attached screenshot shows how two Emacs instances (GUI and
non-GUI) show the contents of a test file (previously written by me
and saved to disk).

IIUC, this time the input method is irrelevant, since the text has
been read from a file (not the keyboard).

As you can see in the modelines, both instances of Emacs have selected
the latin-1 coding system, but the non-GUI instance fails to show the
characters correctly (and not only "á" and "é").

I don't know if this problem has the same root than the original.  If
not, I could file a separate bug report.

-- 
Dani Moncayo