I am having a very similar issue as I prepare to migrate to a new server. Trying to do some testing I do basically a select top 10000 * with an order by date. The point of this was to test tempdb. The performance on the new server is much slower despite it being much more powerful on san disk and server hardware. Not to mention the old server has a heavy production load on it. The query plan is very different with the Sort costing 68% on the new server. Disk latencies in tempdb hit 2000 ms as measured in the OS. As measured on the SAN all are less than 3 ms. The databases are identical as I restored production from backup. Not counting the different query plan it seems there is a bottleneck somewhere since the SAN reports no latency. Have looked at the HBA Que depth settings and made changes with little results. I would like to solve the execution plan issue to know i am comparing apples to apples. Tried moving the tempdb files from the tempdb luns to the DATA luns and had the same results. The data luns have 75 15K spindles under then.

Anyway Ill be watching this thread to see what you come up with.