There's a bit of a paradox here: a monolithic framework gives you less choice because you can't use only a part of it without great pain, however you have ostensibly more choice of functionality from third-parties since there's a standard way to use it that's hard to diverge from.

But if you're any NIH on the side, the framework-as-bunch-of-easily-integrated-libraries approach really makes more sense. It also lets you try a framework incrementally, therefore decreasing risk instead of committing to a behemoth framework prematurely.