> First issue is the one explained by Ann: the engine always prefers

two

> sequential (i.e. storage order) scans for a sort instead of a mixed
> sequential / random approach. This is by design and it cannot be

turned

> off. The problem you experience is caused by the fact that the sort
> records are stored expanded, while they're compressed on data pages.
> This is why you see the big I/O difference. This could be improved

via a

> proper tuning of the SortMemUpperLimit / TempCacheLimit parameters,
> provided that you have plenty of RAM on board. This is a design
> limitation and not a bug.

gotcha. was chewing on Ann's hint about compression while you were
writing:)