I can confirm that the map-reduce thing does not hold up to any performance-wise expectations.
Now that 2.4 is out I'm interested in seeing how much the switch to multithreaded V8 has improved that. http://docs.mongodb.org/manual/release-notes/2.4-javascript/
More seriously though running queries against a master node will regularly (maybe one in two queries) cause it to failover and elect a new primary. You'd think you should be able to run the queries against a slave, which you sort of can. Unless the result set is bigger than the maximum document size (16megs), in which case you're limited to the master as you can't write to a collection on a slave.
We're using Mongo fairly successfully but it has a lot of issues, particularly around administration tasks. Map reduce work gets done in Hadoop.