Understanding STDisjoint() and STIntersects() in SQL Server

The query cost is reported in execution plans as "estimated subtree cost". This is an absolute figure such as 1.5. Conor Cunningham mentioned in a SQLBits presentation that it originally referred to the number of seconds taken to execute on a particular Microsoft employee's machine ("Nick's Machine") in the SQL Server 7 days

but now should be interpreted as a unitless measure of overall cost.

The execution plan is a tree. Each iterator in the tree is given an estimated CPU cost and an estimated IO cost and these are added together to get the overall cost (the relative weightings can be adjusted with a couple of undocumented DBCC commands). The estimated subtree cost includes the cost for the iterator itself and all its descendants. To see an example of the costing formulas used you can look at this article.

To determine the estimated cost for an entire query in SSMS select the root iterator (e.g. the SELECT iterator) to the left of the graphical plan and look at this metric in the SSMS properties window.

When running multiple queries (whether in the same batch or not) the percentage is calculated by adding up all of these values and calculating the percentage as you would expect.

You should be aware that even in actual execution plans this cost figure is based on estimates and using this to compare the relative merits of two different queries can be woefully wrong in cases where the estimates are inaccurate.

There is a hack you can try: where you create a dummy DB with the same name and size as your old data and log files. Then you have to stop SQL server, replace the new empty files with the old files that have data in them.

Check out this blog link by Paul Randal where the hack is explained in detail. It looks like you have tried the first method mentioned, but read along and try the other methods. Please let us know how it goes. I do not know how well it will work without log files, but it is worth a try.