This may be a bit out there for a "lightweight" library but have you thought of implementing a distributed counting semaphore?

In my scenario, I have a sql azure database with many customers. Customers can have a lot of data.
The .net code is hosted in several azure web instances.

I regularly want to process every customer data. However, I don't want to process all customers at the same time because it would overload the database.
I also don't want to process each customer serially (1 by 1) because it would take too long.
Ideally I would use a distributed semaphore with a count of N to process N customers at a time...

It could be built on top of SqlDistributedLock and I wonder if you've come across this scenario before?
Do you see that fitting within the vision of your library?