There are reasons to add and reasons not to add. If there is a niche that is not filled by anything currently, then you should add. Before you do so, though, ask whether it's truly a base concept or a development concept. The Marshal (MH) is a prime example of this. I love the idea of the Marshal, I just don't see a 1st level Marshal -- it's almost a non-sequiter. The Paladin is another (touchy) base class that should be a PrC.
On the other hand, you can use the introduction of a base class to emphasize or promote a sub-archetype within one of the other classes. The Swashbuckler is an example of this in my campaign. I
really like the idea of the lightly armored swordsman. The Swashbuckler class nicely accents that there is a substantive difference between the light fighter and the more dedicated Fighter. Either one is equally dangerous, but there are differences that go beyond which weapon they choose.
A core example is the Ranger. Really, most basic concepts for the Ranger class could be built from some combination of two of Barbarian, Druid, Fighter, Rogue. Having the Ranger class helps to emphasize that there is something more to the Ranger than simply cobbling together other disparate skill sets.
That said, I think it's important to not glut the system with too many classes, though. My own campaign runs with:
- Artifacer
- Barbarian
- Bard
- Druid
- Fighter
- Noble (hybrid of Star Wars and DMG Aristocrat)
- Priest (cribbed from Hong, but slightly modified)
- Ranger
- Rogue
- Spirit Shaman
- Swashbuckler
- Wizard
As far as actual core 4E PH classes, I'd probably keep the Noble and the Swashbuckler from the above list. I could see an argument for the Hexblade, too. Artifacer really doesn't fit with baseline D&D, nor does Spirit Shaman. Paladin, as I said, should be turned into a PrC. Monk fits some games, but not others; take a poll of players and run with the majority. Oh, I'd keep Sorcerer, too -- the only reason it's not on my list is that the spell point system we use really makes the gap between Sorcerer and Wizard too narrow.