The point is that you
can't know what it sounds like unless they are accurate. Every consumer speaker sounds different in different ways. So, if I bought a pair of mid-range car speakers to mix on, assuming most of my target audience would hear it on their car stereo, and I made it sound good on those speakers...even if
everybody only listened on car speakers, odds are good it would still be too hyped or too bassy or too tinny on about half of all speakers. The degree of variance is the problem.
If I mix on a set of accurate studio monitors (and assuming a reasonable level of skill in the area), the sound will be pretty well balanced across any number of speakers. Some will be hyped (smiley curve EQ), some will be bassy, and some will be bright. But a balanced mix will be pretty well-balanced on any set of speakers of reasonable quality. If I mix on speakers that lead me to boost or lower a specific set of frequencies, the speakers that do the opposite will be twice as far off as if I had mixed to a balanced set of speakers.
Everybody here is probably pretty mathematical, so think of a single sine wave at a specific frequency. Say I'm aiming for an amplitude of 1 (units irrelevant). My mixing speaker boosts it by .5, because it's a speaker aimed at guys who like their cars to rattle. So, now I've mixed this sine wave at an amplitude of .66. If someone who likes bright sounding music (and with speakers that drop the bass by .5) listens, my mix now has the bass at .33 (ignoring all the dB math, and the crazy physics of ears/distance/frequencies etc. for the sake of simplicity).
If, on the other hand, I mixed on a speaker that accurately gave me a 1 when I wanted a 1, it'll play at 1.5 on the bass-heavy speakers, and at .5 on the treble-heavy speakers. Smaller differences are better.
Note also that professional studios also tend to include a few sets of speakers. You mix on the "mains" which are usually high quality near-field monitors, and you test on the small and big speakers and possibly a specific set of speakers that matches your target audience. Good engineers also usually live and breathe their projects, and so they'll also listen in the car on the way home and to work in the morning. The main work of mixing, however, always happens on a good, accurate set of monitors. If it doesn't, odds are very good it'll sound like crap on a good portion of speakers.
And to answer this questions specifically:
then shouldn't it sound best when replayed on monitors?
Sure, but how many people have $1000+ speaker/amp combos in their car, iPod, home, etc.? The problem is merely one of making the best mix for the most people. If you know exactly what speakers your customer is using, and those are the only speakers that will be used for playback of your recording, then you would obviously tune your mix for that exact set of speakers. But, that's pretty much never the case. We do the best we can with a complex set of variables.