Improving CLR performance in SQL Server (redux)

Question

I have been spending a lot of time trying to eek out the maximum performance out of a C# CLR UDF. I have already set IsDeterministic and IsPrecise to true, as well as SystemDataAccessKind.None and DataAccessKind.None.
I am now experimenting with the overhead of transferring to CLR.  I created a simple CLR UDF that just returns the input value, e.g.,
[Microsoft.SqlServer.Server.SqlFunction(IsDeterministic=true, IsPrecise=true)]
public static SqlString MyUDF(SqlString data)
return data;
Defined as:
CREATE FUNCTION dbo.MyUDF(@data nvarchar(4000)) RETURNS nvarchar(4000) WITH EXECUTE AS CALLER
AS EXTERNAL NAME [MyAssembly].[UserDefinedFunctions].[MyUDF];
I then use the UDF in a View on the Primary Key (nvarchar) of a table with about 6M rows.
I know there is a small overhead going through a View versus a Table. However, when I query through the table, it is about 2000% faster than querying through the View with the CLR UDF.  E.g., 3 seconds for the table and 60 seconds for the view! I checked
the Query Plans for each and they are both using Parallelization.
I have to assume that all the overhead is in the transition to CLR.  Is that much overhead to be expected?  Is there any way to improve that?
Incidentally, this is a followup to this question:
http://stackoverflow.com/questions/24722708/sql-server-clr-udf-parallelism-redux

Answer 1

Assuming that a way is found to reduce this apparent overhead, what is the intended operation within the function? I ask because the advantages of SqlChars over SqlString might be moot if you will need to operate on the full string all
at once as opposed to reading it as a stream of characters.
Also, with regards to why the CLR UDF is so much faster than the T-SQL version, some amount of it certainly could be the ability to participate in a Parallel plan, but also a change was made in SQL Server 2012 that improved performance of deterministic CLR
functions:
Behavior Changes to Database Engine Features in SQL Server 2012
      Constant Folding for CLR User-Defined Functions and Methods
      In SQL Server 2012, the following user-defined CLR objects are now foldable:
Deterministic scalar-valued CLR user-defined functions.
Deterministic methods of CLR user-defined types.
      This improvement seeks to enhance performance when these functions or methods are called more than once with the same arguments.
Also, 60 seconds down to 3 seconds is a 95% improvement, not 2000%.  Or you could say that the operation is 20x faster without the UDF.
Now, outside of that, I recall seeing in another forum that someone was converting their string to VARBINARY using SqlBinary / SqlBytes and then returning VARBINARY and converting it back in T-SQL. Might be worth a test.