I've noticed that in the proxied poco classes, an entity is marked as modified whenever a property is set, regardless of whether the value has actually changed. I think it would be nicer if only truly modified entities are flagged as modified.

Create a client side profiler that will log running queries and statistics for the various data contexts. Also provide helpful tips to improve performance if applicable. Similar to what NHProf already does for NHibernate.

Change Data Capture is one of the coolest features in SQL Server 2008+. I know it's not core to tailor things for one database technology, but this feature is completely unsupportable in the current EF because EF doesn't allow tables without PKs (and the tables SQL Server gens for CDC have no PKs). Ideally, some sort of real support would be allowed that would allow for some sort of Entity.CDC or something that would string together the tables that an entity uses ... but that is probably a bridge too far.\

It should be great to have a designer which supports all mapping (so which allow us to realize TPC, Horizontal Entity Splitting).
It should be great to have a designer that supports partial model views as I did (http://msmvps.com/blogs/matthieu/archive/2009/04/07/edm-designer-new-version.aspx, http://msmvps.com/blogs/matthieu/archive/2009/04/11/edm-designer-new-feature.aspx)
It should be great to be able to see complex types on the designer and not only on Model Browser window as I did (http://msmvps.com/blogs/matthieu/archive/2009/01/07/edm-designer.aspx).
It should be great to not lose all our ssdl changes when we update the model from db.

I should be able to put basic data validation on all of my public fields.

There are many database tables that have tables with a primary key of type int and are not autoincremented
it should be nice to have Ef generate code to allow updating the primary key.

We have seen that there are lots of discussion surrounding the generated SQL not being optimized. Entity Framework should have a provision wherein User should be able to change the Generated SQL and use it for further processing through some mechanism

There should be an AssociationChanging in System.Data.Objects.DataClasses.RelatedEnd class, so I can validate before items get attached to a relation.

There should be a standard EF Test Suite. As the test suite ideally could run against any of the databases, it would ease out the work required to see compliance against different Databases

Support for mapping System.IO.Stream to Binary fields

It is not currently possible to write a LINQ query that compares a value to data stored in a varbinary column. By comparison, I mean SQL-supported <, <=, >, >= comparisons that lexicographically compare the bytes in the column.

As far as I know there is no way to write these queries without stepping down into Entity SQL.

see http://social.msdn.microsoft.com/Forums/en-US/adodotnetentityframework/thread/f90dec50-acc7-4157-8927-acb246ec3572 for more.

Ability to save contents of a loaded entity set into an xml file to persist into hard disk then load again like is possible in data tables

Some customers want to write queries that include such expressions as "date1 - date2" or "datetime3 + timespan1 > datetime4". LINQ to Entities currently support some of date/time operations through the EntityFunctions.AddX and DiffX functions, but it doesn't support built-in operators with data/time types.

SQL Server schemas (or owners) should reflect into CLR namspace branches or inner classes.

If you have Sales.Customer and Orders.Customer on the database, that shouldn't be Customer and Customer1.

In my models, there are hierarchies that are persisted using TPT while others are persisted using TPH. It would be great if we could choose the strategy and then have Generate Database From Model create the tables depending on our choice.

I would like to see the ability to persist an entire object graph in one DB call. Currently it is way to chatty with the DB which makes it really slow to persist.

Support all mapping inheritance strategies, well documented, as mentioned here: http://www.agiledata.org/essays/mappingObjects.html

Generate the Association objects in a model based on the Foriegn Key Column Name rather than "Association1", "Association2" etc