I don't think it's necessarily a result of messed-up math, depending on your definition of messed-up.
It appears that world-gen basically inputs the available weapons, armor, materials, and built-up stats of individuals into the battle sim...
Much less is going on than you think in the above quote. The math is rather more simplistic and prone to the results in this thread. It does seem to factor in overall experience of the armies, perhaps individually or more likely as a group average, but I doubt it's going down to materials, armor, wound status, etc.
I just generated a world in which dwarves had their usual equipment, then re-generated it without all their armour and all weapons except spears... so, lots of stabbing? No. They were still smashing, shooting, tearing away, etc. They did pretty much the same in battle as well, although I got different civs on the same map each time despite the same history seed.
So, they didn't have mail or any weapons but spears... and yet were doing as well as before, and even giving messages appropriate to different weapon types. I think the weapon messages are randomly generated without actually refering to any equipped status. In fact, the same dwarf can shoot, smash, and stab all in the same battle.
A human, on the other hand, has his hand torn off and goes on to kill 129 more people. Does combat even roll attacks between individuals? Maybe it just takes the average skill levels for the groups, and weights the amount of casualties it generates for a given battle, then assigns names as needed to fill out the casualty reports.
There's much less there than you think. It's mostly random messages painted over a simplistic algorithm. I call it "messed up" because kill ratios are usually obscenely high for city defenders, and skirmishes in the open tend to be even trades between novice groups, even if one side outnumbers the other overwhelmingly, like 50 or more to one.
If dwarven defenders fight their first battle in a mountainhome, as complete novices, the results are fairly predictably the following:
If they are outnumbered 450-to-one, each will generally kill at a 110-to-one ratio.
If they are outnumbered 250-to-one, each will generally kill at a 70-to-one ratio.
If they are outnumbered 150-to-one, each will generally kill at a 43-to-one ratio.
If they are outnumbered 125-to-one, each will generally kill at a 35-to-one ratio
If they are outnumbered 97-to-one, each will generally kill at a 24-to-one ratio.
If a battle is a meeting engagement between dwarves and goblins in open ground, even if the goblins are 9000, the dwarves and goblins will kill similar numbers (a dozen or so, at 18 dwarves) and retreat. This was a constant no matter how large the goblin army.
Try generating different populations of dwarves and goblins. I've done from 9000 to 1000 goblins and 18 dwarves with the above results a dozen times each. The game favors the defenders particularly the more they are outnumbered, counter-intuitively. Meeting engagements in open ground between imbalanced numbers usually end up with a few of each dying, no matter how overwhelmed. These and other aspects lead to a reasonable definition of "messed up."
The question is, then, if world-gen is messed-up, what should happen? If the last survivor is indeed bad-ass, why shouldn't he be allowed to survive? And, is there anything wrong in discussing the epic legends that the DF world-gen creates?
Even novices can kill at high ratios. in fact, the more outnumbered, the more they kill. Bad algorithm, IMO. Discuss if you like, though.