First, don't mix in headphones.  I do, and others do, but I've
been mixing for decades (amost 3 now!) and I know what the
headphones do, and I try to compensate.  For me, it's due to
a temporary restriction mainly based on decorating concerns
(my studio nook is "temporarily" in our family room) and to
be able to work at night while others are asleep.  But I do
NOT reccommend it -- even for an experienced pro.  It's a
compromise I'm willing to make as an amateur.  I'm willing
to settle for less than my very best -- things still turn out
good enough for me.  (You can hear my work at Learjeff.com.
 Folks who know their stuff will detect some problems that
were probably caused by my monitoring!  But my friends &
family don't seem to notice.)

The ideal mixing level with respect to EQ curves is 87 dBSPL(C).
Now, what the heck does THAT mean?  It means, hop on down to
Radio Shack, get a Sound Pressure Level (SPL) meter.  I like
the cheapie analog one for $30 best.  Slap that puppy on some
kind of stand (IIRC, it has a camera stand mount), put it where
your head is when mixing, and back away a couple feet.  While
playing commercial music, turn up your monitors until it reads
between 85 and 90 dB, with the switch set to "C" weighting
(for music -- "A" is for speech).

This is the ideal listening level for EQ and best sensitivity.
HOWEVER, I recommend mixing at lower levels (though not to
"nearly inaudible" levels!)  But before lowering the level, we
have to consider Fletcher Munsen as mentioned above.  If you're
using a home stereo amp for your monitors and it has a "loudness"
switch, turn that ON and forget about Fletcher Munson.  If not,
you'll want to manually compensate.  Post back about that, maybe
I can write something up after a little study.

Now, assuming you have Fletcher Munsen covered, I recommend mixing
at significantly lower levels, like 60 to 70dB SPL(C), which is
conversational level.  (It really helps to have a quiet room.)
But, while I recommend MIXING at this level, you also have to
double-check your mixes periodically at higher levels -- ideally, 87dB.

Here's why:

1) reduce ear fatigue, as you know

2) When mixing loud, you hear more detail.  But don't you want your
   mixes to sound good when played quietly?  I sure do!  I often
   used to find after mixing loud, that an important underlying part
   was dropping out entirely when turned down.  I don't want this to
   happen!  Mixing quietly is harder, because it forces you to do
   whatever's necessary so that all the important elements are heard,
   even when your ear isn't getting all the detail it would in a loud mix.

3) I find that mixes made loud generally sound bad when played quietly,
   and mixes made quiet generally kick ass when cranked up.  Of course,
   there can be some nasty artifacts in some plugins or tracks, so
   you MUST periodically CHECK your mixes and overall EQ at loud levels.
   Just don't work that way as a habit.

I hope this clears up your doubts.  Your questions make perfect sense;
this isn't a cut-and-dried simple matter.

Where does the 87dB number come from?  I got that from Bob Katz.  He got
it from motion picture sound engineers (SMPTE), where it's used as the
standard level for movie audio, based on a good deal of research. 
Katz did some informal testing to verify that it was good for audio
music mastering as well. 

TomS, you're right that we do tend to adjust.

So, if you always listen at 70dB SPL, and listen to all your
favorite CDs at this level, and mix at this level to make your
mixes balance out similarly to your favorite CDs in the same
genre, then things will work just fine.  Loudness compensation
is more important if you listen at different levels at different
times.  Even then your ear (really, your mind and ear) adjust
somewhat, but they don't adjust completely.

It is the kind of thing where you can use your judgement rather
than calibrating your system and using the Loudness switch, etc.
Being aware of it, and listening to the effect in well-engineered
commercial CDs, and just [i]paying attention, using your ears,
and using your brain[/i], you can get by quite well without
perfection.

But I suggest you use that loudness switch, if you have one.
Some audiophile purists would disagree, because it (any filter)
causes some phase shifts.  And they'd have a good point, but
my mixes (and most mixes done by amateurs) won't suffer too
much from the small amount of artifacts introduced.  But
that's JMHO.  Folks with more sensitive ears might know better
than I do!  I'm definitely not a golden-eared one.

Hey, fvfv: something that's a lot more important than Fletcher
Munson compensation, and more important in the short term to
listening levels: speaker placement.  Sure, you're using
headphones now.  But when you move up to speakers:

1)  Don't use the surround-sound system -- not made for this task!
    They're made to make "everything sound lush and wonderful", a
    lot like many Bose speakers.  Not good for mixing.  Maybe if you
    have a really good system with fairly flat speakers, and you
    disable the rear speakers for the task, OK.  But by "really good"
    I mean audiophile stuff, not home-theater-geek stuff.  You CAN
    use that system for "comparison monitoring", when you check an
    almost done mix on everyone's stereo (bring the beers and make
    the rounds!)

2)  What you want are relatively flat speakers, but don't need
    to be fantastic fabulous wondrous etc.  The key is: you can't
    fix what you can't hear.  Relatively small speakers are best,
    using "near-field monitoring".

3)  Place the speakers at roughly ear height, in an equilateral
    triangle with your head at one corner and the speakers at the
    other two.  Same distance between all.  (You can depart from
    this somewhat, up to a 45 degree angle from your head.)  They
    should be from 3 to 6 feet away, ideally more like 3 or 4.

4)  KEEP THE SPEAKERS AWAY FROM THE WALLS!  Ideally, as far from
    the wall as from your head.  Practically, the first foot is
    crucial, 2 feet away really helps a lot, and after 3 feet you're
    starting to hit diminishing returns.  This helps to take the
    room out of the equation.  It doesn't do it fully, but it does
    take the 1st order effects out (direct reflections, but ignoring
    resonance).

5)  Listen to a LOT of your favorite music in your mixing gear,
    just the way you'd listen to your mixes.  Really.  A LOT.  This
    really helps to avoid the need to compensate for Fletcher Munson
    and friends.  Feel free to EQ your monitor setup a little so
    that it sounds pleasing to you, but it's probably best to get
    real used to it run flat first, and only adjust much later if
    you really feel it's needed.