Here are three simple tests that read the same data, yet report very different logical reads:

### Setup

The following script creates a test table with 100 identical rows, each containing an *xml* column with enough data to ensure it is stored off row. In my test database, the length of the *xml* generated is 20,204 bytes for each row.

    -- Conditional drop
    IF OBJECT_ID(N'dbo.XMLTest', N'U') IS NOT NULL
        DROP TABLE dbo.XMLTest;
    GO
    -- Create test table
    CREATE TABLE dbo.XMLTest
    (
        ID integer IDENTITY PRIMARY KEY,
        X xml NULL
    );
    GO
    -- Add 100 wide xml rows
    DECLARE @X xml;
    
    SET @X =
    (
        SELECT TOP (100) *
        FROM  sys.columns AS C
        FOR XML 
            PATH ('row'),
            ROOT ('root'),
            TYPE
    );
    
    INSERT dbo.XMLTest
        (X)
    SELECT TOP (100)
        @X
    FROM  sys.columns AS C;
    
    -- Flush dirty buffers
    CHECKPOINT;

### Tests

The following three tests read the *xml* column with:

1. A plain `SELECT` statement
2. Assigning the *xml* to a variable
3. Using `SELECT INTO` to create a temporary table

```
-- No row count messages or graphical plan
-- Show I/O statistics
SET NOCOUNT ON;
SET STATISTICS XML OFF;
SET STATISTICS IO ON;
GO
PRINT CHAR(10) + '=== Plain SELECT ===='
    
DBCC DROPCLEANBUFFERS WITH NO_INFOMSGS;
    
SELECT XT.X 
FROM dbo.XMLTest AS XT;
GO
PRINT CHAR(10) + '=== Assign to a variable ===='
    
DBCC DROPCLEANBUFFERS WITH NO_INFOMSGS;
    
DECLARE @X xml;
    
SELECT
    @X = XT.X
FROM dbo.XMLTest AS XT;
GO
PRINT CHAR(10) + '=== SELECT INTO ===='
    
IF OBJECT_ID(N'tempdb..#T', N'U') IS NOT NULL
    DROP TABLE #T;
    
DBCC DROPCLEANBUFFERS WITH NO_INFOMSGS;
    
SELECT 
    XT.X
INTO #T
FROM dbo.XMLTest AS XT
GO
SET STATISTICS IO OFF;
```

### Results

The output is:

```none
=== Plain SELECT ====
Table 'XMLTest'. Scan count 1, logical reads 3, physical reads 1, read-ahead reads 0, 
    lob logical reads 795, lob physical reads 37, lob read-ahead reads 796.

=== Assign to a variable ====
Table 'XMLTest'. Scan count 1, logical reads 3, physical reads 1, read-ahead reads 0, 
    lob logical reads 0, lob physical reads 0, lob read-ahead reads 0.

=== SELECT INTO ====
Table 'XMLTest'. Scan count 1, logical reads 3, physical reads 1, read-ahead reads 0, 
    lob logical reads 300, lob physical reads 37, lob read-ahead reads 400.
```

### Questions

* Why are the LOB reads so different?
* Surely the exact same data was read in each test?