A month or so back, I posted a barbarian class (later renamed berserker) to the OD&D board (you can read the thread here). It brought up a couple of points that I think are relevant.
First - there was a bit of backlash to the merging of the barbarian and berserker archetypes that I had perpetuated in the original sketch of the class. The separation of the two is, I believe, a bit artificial - Conan did have moments that are best described as berskerk rages. And it's a very old conflation in the hobby. Issues of Alarums & Excursions had a berserking barbarian as far back as 1976, and the 1977 Arduin Grimoire codified it as a class. So I don't feel that I was entirely off base. But, there are legitimate differences at work here, and since the "berserker" is in Monsters & Treasure, I changed the name to reflect this.
What I find equally important is the question of "What should be a class?". I view D&D as best played within a dungeoneering context, which plays a significant role here. The major classes that people were interested in during the early period of D&D were the paladin, the ranger, the bard (or singer or poet), the druid (or neutral cleric) and the barbarian. Most of these are outdoorsy types who are not necessarily a natural fit for the dungeon. I play the game with the Greyhawk paladin (pretty much any Charisma 17 character becomes a paladin), and I'm considering the druid as an addition.
I've been reading some more of the Conan stories lately, and Conan is plainly a fighting man. He can do a lot -- definitely at different points he's a hero and a super-hero -- but class wise, I would not put him in my berserker class. There's a certain cachet to being a barbarian, but there's not a lot of mechanical flavor that is going to differentiate the character from the OD&D fighting-man without going the weird route pursued by Unearthed Arcana or merging with the berserker. And I'm starting to think that this is okay.
The niche that I think has not been filled adequately is the lightly armored, clever type who dabbles in magic but isn't serious about it, and is a hell of a fighter nonetheless. (The Gray Mouser, Cugel the Clever, etc.) There were echoes of it in the Greyhawk thief class, but this was merged with a specialist who is probably best left as an NPC. I think that some adaptations of the bard came closer, but really it's one of the challenges that remain, and something I've spent some time thinking about.
Beyond the examples I've listed above, I think it's necessary to have a certain skepticism about the need for a new class. For instance, I've thought about a more potion-oriented magic using class as a witch, but too much of the interesting detail would happen outside of the game. It's a valid character type in the world, but I don't know if it translates to an interesting D&D class.
I think that this approach needs to inform our class-building. There's a certain degree of flexibility within the three original classes, but they tend to move in definite ways in-game. (For instance, fighters tend toward AC 2, magic-users go from one-shot "sleep 'em!" to world-shaking magic, etc.) But classes still need to be added with a "should I add this?" approach rather than "this would be cool."