I was looking into it as a scalable solution for the backend of an app, but I was turned off by reports of high latency (up to a couple seconds) and the initially low limits for the service (they've been upped, but I think it was 60 second timeout per lambda and max 100 concurrent lambdas across an entire account; now it's 300 seconds and 1024 concurrent processes).
Lambda seems designed for internal trusted use for offline or background data processing, not for writing a public API endpoint.