Celebrim
Legend
I agree with the OP that free multiclassing defeats the purpose of a class-based system.
[MENTION=4937]Celebrim[/MENTION] listed five possible reasons for using classes. IMO each of them is defeated by being able to multiclass freely.
Well, despite the fact I'm mentioned, I don't agree.
"a) Easy to grasp archetypes or roles." - If I can multiclass freely, at some point the resulting characters won't resemble easy to grasp archetypes or roles. However, this doesn't defeat the original purpose. Nothing is forcing people to multiclass, and the easy to grasp easily entered archetypes and roles are still there and still available to those that want them. Multiclassing is a more advanced player behavior, and presumably those that attempt it do so only after having gained some experience with the system.
"b) High balance between classes." - If I can multiclass freely, the result is by no means guaranteed to be markedly weaker or stronger than single-classing assuming we've a reasonably well constructed system. If you feel otherwise, then let me suggest that your experience with multiclassing is being unfairly colored by experience with 3.5 D&D. However, if you examine the problem I think you'll find that it's not multiclassing that breaks the balance between classes, but two things - low balance between spellcasting and non-spellcasting classes, and prestige classes that were unfortunately never designed to be balanced in the first place. The big problems in 3.5 were things like poor play testing at high levels of the initial system, PrCs, and poor overall playtesting of splat books as they were introduced. None of that had anything to do with multi-classing, as classes like Druid was broken even solo-classed in 3.5.
"c) Enforced character breadth." - If I multiclass a class with high enforced character breadth to another with high enforced character breadth, the result is still high enforced character breadth. The culprit in 3.5 is once again - Prestige Classes. The huge amounts of PrCs that were published to generate sales of books meant that for any given concept there were usually 3-5 variant PrCs that overlapped the concept. So sure, under such circumstances you dip 3-5 classes with overlapping theme to produce very narrow characters with single really powerful tricks. But that's not a problem with multiclassing itself, but rather with the PrC concept combined with conscious willful destruction of game balance by the publishers for the sake of selling books.
"d) Regular predictable progression of abilities." - If I multiclass in the 3.X style, then I'm still a 5th level character regardless of what classes I took. I have a predictable number of HD and predictable maximum abilities, and a predictable range of attack bonuses, defenses, skill ranks, and so forth. Therefore the original goal is not defeated. This is actually a bigger problem in 1e style multiclassing/dual classing, which produced highly unpredictable combinations of abilities and levels for a given amount of XP.
"e) Easy mix and match to the desired setting." - If it was easy to select a subset of archetypes available to characters in a setting, then presumably the combinations of these roles to a large extent also fit the setting. Afterall, even in a setting with well defined roles, not everyone is going to have a simple life story but could be forced through circumstance, luck, divine intervention, and sudden change of social class to change roles.