Given their reaction, it would be nice if they were more upfront and honest about it on their website. A lot of people -- including myself -- were bitten by the unexpected. Less issues with durability than performance, of which we had very high hopes considering how much durability is given up.
Specifically, it would have been nice if they outlined the known issues surrounding compaction and how much of an impact not running compaction will have after weeks or months.
Essentially, almost every reasonable use case absolutely requires either taking a regular outage for compaction or using a 3-node replica set to do a rotate-compaction dance. This is an EXTREMELY frustrating thing to run into and surprises like this that pop up after deployment are unlikely to earn them any allies.
10gen has been publicly talking about working on online, incremental compaction for over a year now. Riak, CouchDB, Cassandra, Redis (AOF), and HBase all have incremental, online compaction.