But the 3.x paradigm is, rolling a new character is a major league chore and you can't have hangers-on because then (heaven forbid) one character might be more powerful than another (shock, horror).
Rolling a character does involve a lot of options. This points to a 3e encouraging a more narrative style of play than 1e, something it probably inherited from 2e. The idea is more that epics like Lord of the Rings didn't kill the main characters, so if the main characters can keep coming back, this preserves some semblance of narrative, as well as letting you hold onto a character you like.
The henchman issue is more to do with 3e encouraging very strategic combat (often by ways of minis). A lot of troops = a lot of strategic elements.
So, compared to 1e, 3e has more options and more character consistency, at the expense of some ability to do it fast, and to do it with bulk numbers. IMO, this is a good trade -- I prefer character options and narrative structure over exactly figuring out the stats for 30 goblins and rolling combat against them and 20 redshirts. I'd rather invest my hour in one character than a dozen. It's not for everyone, but it works for me.
Thus the logic that allows characters to be raised an infinite number of times in the core rules, and thus the problems well illustrated by this thread.
I think that throwing out character options are baby-with-the-bathwater material. With mechanics like action points you don't even throw out death per se, you just make it astronomically rare for PC's (which further supports the idea that 3e characters should be consistent from session to session). By and large, that's good enough for verisimilitude. "PC's are special, so they avoid death more often" is a perfect narrative-game device.
Really, there are kind of two different trains of thought running through this. One is how rare death is: rarer death is more significant, which adds some great drama. Common death is more "realistic," which does make D&D feel more like a game and less like a novel (definitely appealing). The other is how powerful death is: a death that is a speedbump would rob a rare death of it's significance, but it would be perfect in a game-centric kind of etting where adventurers die in droves. Inalterable death means that there's some pathos, but it gets frustrating if you're hitting it over and over again -- the pathos gets numbing.
NPC-replacement is "realistic," but it also is speedbumping: 10 minutes later, you have a new character (not unlike standard 3e's resurrection, but instead of fixing the mask, you just get a new mask).
Action points are more dramatic, but it tends to go hand-in-hand with rare resurrection, which means that death is there only when it's really dramatic. This doesn't fit with 3e's standard model, but fits with Eberron's "mass magic" and "adventure movie" feel.
It's a major argument in favour of a rules-lite approach, isn't it?
Not really. The convo (or at least the OP), AFAICT, isn't about how to make generating a character or replacement character simpler, but rather about how to avoid having to generate new or replacement characters at all.
There's a lot of good arguments for rules-lite, but death isn't pervasive. You can have rules-light and rare-death just as easily as you can have rules-light and frequent-death. It's just a problem when you have rules-heavy and frequent-death. If your problem isn't the rules, you just need to fix the death (which is pretty easy to do without losing it's punch from a storytelling perspective).