For startups relying on Mongodb for critical infrastructure, the cost of having a poorly implemented mongodb cluster can actually be high enough from a morale and product performance standpoint to ruin the company. I've seen it happen a few times to very well funded companies where their inability to quickly solve MongoDB issues made it impossible to continue iterating on their product, and eventually caused most of their core backend team members to quit.

So I think it's super important to de-risk MongoDB as quickly as possible and then move on with life. If you can use bare metal, just switch to using SSDs or even FusionIO cards. The new AWS SSD instance types should also be ok. Pay the money you need to service providers like AWS, Rackspace or Softlayer etc. to get it done.

In a best-case situation, you can keep upgrading every 12-18 months and rely on Moore's law to allow your MongoDB machines to scale with your growth. But even in a bad situation where your database is growing in size too fast, this vertical scaling strategy should still give you 6-12 months of breathing room, to allow your dev team to cleanly switch over to a different database.

I see a lot of teams starting out with MongoDB, but the smart ones know it's a big risk and once they find market fit, their first major initiative is to switch over the database backend to use something like Cassandra + Mysql.