but the generated code is often not as optimized as what many would write. For instance, even if you're aggregating multiple columns from a single child table, it uses correlated subqueries quite heavily and inefficiently.

We must be doing different stuff- I was quite impressed by the way aggregates are handled. So can you please provide examples of the inefficient subqueries of which you complain?

As for consensus- people form consensus about lots of things. Most of us expected that WMD would be found in Iraq, for example. In the case of Linq we do have the ability to check the facts for ourselves before forming an opinion, so that's what we should do. IMHO.