Currently, there is a global read/write lock. MongoDB does extensive yielding and tracking of on vs. off disk to mitigate the effects of this. In subsequent versions (2.2) , we'll be lowering the scope of this lock as well.
http://www.mongodb.org/display/DOCS/How+does+concurrency+wor...
Note, even with Read-Slaves, you are STILL blocked, since each write replicates to each Read-Slave, and counts as a blocking-write on that slave while it replicates in.
We dramatically increased application performance when we refactored to remove a lot of little writes for things like caching, and moved them to memcache.
This is even more true for mongo than for other databases because mongo doesn't degrade very well; when you overload it with writes it will effectively grind to a halt.
This is documented e.g. in the clustrix benchmark: http://sergei.clustrix.com/2011/01/mongodb-vs-clustrix-compa...