Good essay and interesting points
I have my own issues as well with PrCls, although I indeed use them for PCs and NPCs.
Joshua Dyal said:
Psion said something in another thread, which I'm going to paraphrase, that struck me as particularly insightful though; good prestige classes are much more than simply mechanics, they are character concepts, and that's something that's worthwhile.
According to his review numbers

Psion might have seen a thousand PrCls already... How can one not agree with this statement of his? I think the Shadowdancer is the perfect example about how to write a PrCl: it's a pretty simple idea ("shadows") followed by gathering different mechanics/abilities related to that or adapted to the idea, even if they usually belong to different character archetypes (spells, stealth, summoning...).
It may be a little silly, but I tend to have prejudices against "short" PrCls with 3-5 levels, because they don't usually give me the feeling to have been planned enough by the authors. For the same reason, I look with suspicion at PrCls that give bonus feats (always seems the trick to balance up a PrCl the design of which gave a weak result).
On the other hand, some PrCls don't really follow a clear concept but nevertheless are very good IMHO. Those are the PrCl that lack a progression but instead give one specific ability by choice every level: Archmage, Hierophant, Horizon Walker. In this case it doesn't really matter how many levels they have. They don't really have a concept (yep, "very powerful wizard with unique various abilities" cannot really qualify as a concept guys...

) but they are instead used
exactly as an alternative game mechanics. I am in favor of these PrCl, as long as they are don't have balance problems.
Joshua Dyal said:
The theory of prestige classes allows, I think, for two specific meta-game concepts to be realized:
- Belonging to an organization that grants special class-related benefits (i.e., the Harpers, Red Wizards of Thay, etc.); or,
- Narrowing in on a concept that is too specific to really be a core class
The problem is, the more I've thought about it, the more I think PrCs are suboptimal at both of those tasks, for the most part.
Except the last use of PrCls I have mentioned above, I think you summarize quite well what PrCls are for.
With the first metagame idea I agree with you: existing PrCls are not very well designed for this task. But the way I see it, these are the problems with PrCls and organization membership:
1) Gamers are stuck with the idea that to the only way to be member of an organization, in 3ed you MUST take the signature PrCl.
2) Writers design a PrCl thinking about which class (usually only one) is meant to qualify for it. Requisites usually cut off every other class, either for the other classes it's too expensive to qualify or the PrCl simply demand a class ability which belongs to only 1 core class.
This is actually against the idea originally explained by Monte Cook, that PrCls should be open to everyone (although not just as easy). Unless the whole PrCl is meant to revolve around a specific class feature, the only true reason for writers to restrict a PrCl is that they cannot playtest it with more than 1-2 characters, so it's best to restrict it as much as possible.
3) Requirements should follow the concept and not be used as a design trick to follow point 2) or to make the PC pay a cost for the next abilities, which sucks unbelievably hard...

To me most of the PrCls around have requisites designed around these two points (plus the minimum level) and only a very generic adherence to the concept.
Joshua Dyal said:
And the problem with narrowing in on a concept that is not served by the core classes is, that you want to play that concept from the moment you envison the character; you don't want to "graduate" finally into the concept after levels and levels of play. Granted, you can still roleplay the concept, but if the mechanics don't support you, then that's an unsatisfying experience.
I agree with you that changing details of a core class or using new feats is a much more elegant way to build a concept than make a PrCl.
I really think that if PrCl had better planned requirements, playing the concept since the start would be feasible. Instead you often need to stack feats-skills-spells you wouldn't otherwise bother, only to get into the PrCl ASAP.
However let's not get crazy for "character concepts" either. IMHO it is a vastly abused excuse to crave for cheesiness.

D&D is a game about choices, and of course it would be the best if every choice, every option, every idea behind a character would be equally effective to play. If someone is developing a PC and making suboptimal choices for just a style reason, I would defitely like to help him not lose too much game-wise.
But when you say "if the mechanics don't support you, it's an unsatisfying experience" sounds like we are more focused on a
mechanical concept than a character concept. We are just too much used to thinking that character = set of scores and abilities. An example: I have just argued with someone about a Ninja character, the guy said D&D sucks because no one is able to make a Ninja class, and I told him that there is indeed a Ninja class D&D, and it's even in the PHB!

Actually, there are at least 3 Ninja classes if you want them to be, depending on your character's priority, but I admit the Monk fits better because of the special athletic features. He argued that a Monk doesn't work because it has no sneak attack (he actually says it doesn't work also because it doesn't have full BAB).
Obviously this is a very specific example, in many other cases the core classes don't really allow what one is looking for (undead servant for a necromancer e.g.), but with a little fantasy on the DM's part you can always achieve this sort of things without PrCls...