I just got another crash today and was able to get a bit more diagnostics using IntelliTrace
What I am seeing:
1. The "server busy" state is triggered by a barrage of OData calls (they trigger a bunch of underlying connections to SQL Azure)
2. We already have a throttle in place that stops clients from performing more than 1 request a second
3. Memory seems fine when Azure flips this to busy
4. There are a series of SqlExceptions preceding the fault: "The client was unable to establish a connections because of an error during connection intialization"