I'll have to read up on that. My problem with replica sets is that they serialize all writes through a primary instance of the mongo daemon. That's fine. But the problem is, it seems like the primary can be super-saturated with write requests. This aids in creating replication lag.
And even then, what's the point of "high availability" if like half of all database accesses are writes anyway!
You have serialized writes no matter what but the awfulness that reads may be behind!
What's the point of even horizontally scaling anyway? Not everything should be parallelized!!!!