> (tool-bar-mode -1)
[...]
 >   (setq default-frame-alist
 >         (append (list (cons 'tool-bar-lines 1))
 >                 default-frame-alist))

Inherently, you're painting yourself into a corner by simultaneously
demanding and rejecting toolbars.  Moreover, that last assignment gets
you a second entry for `tool-bar-lines' in `default-frame-alist' (and we
have received _lots_ of mails from you complaining about such duplicate
entries before).  However, you do _not_ set `initial-frame-alist'
simultaneously, so you won't get a toolbar for the first frame which is
the correct behavior since you have toggled tool-bar-mode off before and
only that counts for the initial frame (unless you explicitly override
it).  Since `default-frame-alist' contains a tool-bar-lines entry you
will get a toolbar for future frames which is correct as well.

That said, the behavior of Emacs _is_ wrong because it should not say
that `tool-bar-mode' is on when there are no toolbars.  The reason is
that `tool-bar-mode' does

   (if tool-bar-mode
       (progn
   	(dolist (frame (frame-list))
   	  (if (display-graphic-p frame)
   	      (set-frame-parameter frame 'tool-bar-lines 1)))
	(if (= 1 (length (default-value 'tool-bar-map))) ; not yet setup
	    (tool-bar-setup)))
     (modify-all-frames-parameters (list (cons 'tool-bar-lines 0)))))

so you can easily see that for turning `tool-bar-mode' off, default and
initial frame parameters are set.  For turning it on, only the
parameters of the existing frames are set and _no_ default or initial
parameters.  Since the `display-graphic-p' test is not useful anyway, I
propose the attached patch.

Yours truly, martin.