It is worth noting that Entity Framework has (at least) three ways of being consumed:
- LINQ to Entities over Object Services over Entity Client
- Entity SQL over Object Services over Entity Client
- Entity SQL using Entity Client command objects (most similar to classic ADO.NET)
Entity Client ultimately spits out a representation of the ESQL command (in a canonical, database agnostic form) which the ADO.NET Provider for the specific RDBMS is responsible for converting into store specific SQL. This is the right model IMHO as over the years a lot of time has been invested (and will continue to be invested) in producing great ADO.NET Providers for each store.
As Entity Framework needs to work with many stores and therefore many ADO.NET Providers there is less scope for them to easily optimise what the Entity Client generates on a per store basis (at least - thats where we are with v1). The LINQ to SQL team had a much smaller problem to solve - "works only with SQL Server" and hence could do store specific stuff more easily. I know the EF team are aware that there are cases where EF to SQL Server is producing TSQL less efficiently than L2S and are working on improving this for V2.
Interestingly this model allows new capabilities to be added between the Entity Client and the ADO.NET Provider for a store. These "wrapping providers" can add services such as logging, auditing, security, caching. This is discussed as a V2 feature over at http://blogs.msdn.com/efdesign/archive/2008/07/09/transparent-caching-support-in-the-entity-framework.aspx
If you look therefor at the bigger picture you can see that it would be horribly difficult and indeed restrictive to try and somehow retrofit L2S TSQL generation into the archiecture of the Entity Framework.