Sadly, I don't see any good way to salvage multiclassing in 3E without some serious rewriting to the point of making a new edition.
Just curious, what do you define as good multiclassing? What do you think are the failings of D&D's multiclassing system? (more than 'they aren't productive', like what goals are they missing?)
There are a lot of issues with D&D's system in 3E. In short, you can see this in that pretty much no caster multiclasses (excluding PrCs) and pretty much every mundane character multiclasses. This is because casting is awesome, so you don't want to give that up, but all mundane characters have is BAB, skill points, and class features. Those all stack. So you dumpster dive looking for the best features to stack while keeping your BAB or Sneak Attack or whatever as high as possible.
Balance is another issue. Given that casters are so awesome, it's hard to make a balanced system with that issue looming over head. Since wizards are awesome and fighters are not, you're stuck making a gish either worse than a wizard (so why be one?) or as good as a wizard (in which case, why would you
not be one?). Sure, there are flavor reasons to play a gish, but that's really it. So, this is just one more issue.
There's no easy way to fix this since the stacking aspect of 3E's Hit Dice are built right into the base of the system. So, you get kludges like dual-progression classes (Mystic Theruge, Eldritch Knight, and Arcane Trickster from core), Multiclassing feats (from Complete Adventurer and Scoundrel), and ACFs (like the Battle Sorcerer). There's also the Initiator Level mechanic from ToB, but that has some problems. It's nice in that it basically has a Mystic Theruge-like progression built right in, but it has problems in that it seriously matters what order you take your classes in. From an optimization standpoint, that can be fun. For a DM trying to make an NPC real quick or other forms of quick char-gen, it's a pain in the ass.
I haven't used Incarnum before, so I can't comment on it. I know Frank recently said at the Den that if 4E had used it instead of ToB as a basis for their system, multiclassing would have worked nicely. Perhaps I should look into it.
As for what I'd like in a
good system, I'd have to give that some more thought. I'll be honest that I'm so entrenched in a 3E way of thinking about things that I have a hard time thinking out of the box. 2E had an interesting system that sounds cute in principle, but in practice, it depended on what you picked. It was a pseudo gestalt where you'd pick two (or three) classes and track your level in each of them, but you had to split your XP between each class, so you advanced more slowly. I say pseudo gestalt because you averaged your HP rather than picking the better of the two. In all other ways, you picked the best of each class given each of their levels. You have to remember though, that back in 2E, each class had a different XP progression and that it took double the XP to level each level for a while. So, in practice, if the group was mostly level 4, you'd likely be level 3/3. So this system only works if having two classes at level X-1 is the same as having one at X.
One way to do it might be to simply build in mandatory multiclassing right into the system. You're forced to pick a main class and a sub class, and you get abilities from your sub class (similar to Final Fantasy Tactics). Now, if you reject this outright because there's no straight classing then, another option might be to allow you to pick the same class as your primary and sub, but I'm not sure how it would work out. I've never put any real work into a system like this, so I don't know.
Another option is to go classless. If you want to "fake" classes, you could create a bunch of "classes" which would just be groupings of abilities for quick char-gen and reference, but the idea would be that as you gain levels, you simply pick level-appropirate abilities and call it a day. A serious downside to this approach is balancing that many combinations would be a
nightmare.
