I'm not a 3E player but I don't mind heated discussion.
The following seemed to me some key features of the spell (from the SRD):
Several illusory duplicates of you pop into being
<snip>
The figments stay near you and disappear when struck
<snip>
These figments separate from you and remain in a cluster, each within 5 feet of at least one other figment or you.
<snip>
Enemies attempting to attack you or cast spells at you must select from among indistinguishable targets. Generally, roll randomly to see whether the selected target is real or a figment.
<snip>
While moving, you can merge with and split off from figments so that enemies who have learned which image is real are again confounded.
One thing I notice is that there's no indication as to who decides where the images are located. With 8 images, plus a caster, it seems they could be spread out over a 40' line. But that doesn't really make sense for many attack, especially melee attacks: presumably when a melee attack is made (which is targeting a particular entity in a given 5' square), the images all cluster together about the caster. But this doesn't really make much sense either: a blow against an illusory figment which is one of 8 in a 5' square, plus a caster, is almost certainly going to carry through and take out more figments (or the caster).
So maybe only
one figment joins the caster, in the same square but (say) 2 or 3 feet away. But then the miss chance would only be 50%, not 1/(N+1), where N is the number of images.
My provisional conclusion: this is a spell whose mechanics made sense in AD&D, with abstract positioning and 1 minute rounds, but it really doesn't survive the translation into the much less abstract 3E positioning rules.
(EDIT to help warm things up: can we talk about Mirror Image as a "dissociated mechanic"?)