The workflow for 3D models is most likely less complicated than rendering all your models to sprites. If you change some detail in a texture, you don't have to render all the sprites again etc.
Since you're targeting flash, you'll have to consider the limitations of the platform as well. The new 3D APIs (molehill) are capable of performant 3D graphics, however it might take quite some time till this player will be released and has a wide user-base.
Current flash 3D engines like Papervision or Away3D could work, they will put a heavy load on the CPU though, even with low poly-count. Also z-fighting is a constant issue with these engines. If you plan to use a fixed viewing-angle, you can prepare your models in a way that minimizes these issues though.
If you go the sprite-route, you can leverage existing frameworks like Flixel or Flashpunk and it won't matter how complex your models are. Also the amount of monsters/entities you can display on screen will be much higher with the sprite approach.
There's probably lots of people that play games on Facebook that have an old and/or slow CPU (eg. netbooks), so performance can be an issue.
So for your target audience, it would be best to go the sprite/bitmap route as of now. 3D will become viable with the new flash 3D APIs. If you're interested in that, this project might be a good way to learn about this new technology. Early adoption might give you more visitors, but it's also harder to do since APIs might change or tools/frameworks aren't mature yet.