We are testing mongo for production. Right now we do only write operations, with 50 threads simultaneously. Each document is about 190bytes. We have 2 shards, and each of those two shards is a replica set with 2 members and 1 arbiter. Sharding is based on a UUID so it's pretty random. After the crash had 40280474 documents which results to about 2K inserts per second, perhaps low for this h/w.
After about 5 hours of normal operation, mongos crashed with segfault. All logs are attached. (from mongos and the shards). The shards stayed operational and replica sets reported no errors when connecting to them directly after the crash. The db seems operational after mongos restart. This was our first test.