mrviking
Feedback: Since your intention is to recommend relevant items you might want to optimize the referrer experience. It took a lot of clicks to go selecting a list to accessing the relevant content. E.g it took me 3 clicks to find http://www.listnerd.com/item/savant-1, and then I gave up on finding the artists music or a link to it on that page. On bagcheck it takes 1 click to access the relevant referral link.
It sucks to be the alpha tester. However, the SQL vs NoSQL debate is silly, because Facebook seems to be using both to scale.
A lot of content at Facebook seems to be served from a denormalized NoSQL storage and that is the useful takeaway insight. I do not know what I can learn from the fact that a team with some of the most talented SQL ops and developers can store some core data in SQL.
I can understand that the customer is frustrated with having to change his code to move from a VPS environment to a cloud environment, but that is the unfortunate truth when scaling. Either you 1) get closer to the real hardware and throw resources at it (more ram, vps, or dedicated machine) or 2) change your code to be horizontally scalable on light cloud instances.
Maybe this is chance to create a help-page that explains to users what resource limits they should stay below (memory, cpu, queries) to handle peaks in traffic and how to identify problem points (bad SQL queries, hungry plugins).