69.5% of the total batch cost for the derived table method according to the execution plan, but only 31.8% of the runtime (according to the getdate() queries I inserted at various steps)... I'm confused. I'd assumed that cost and runtime were directly correlated unless there were competing processes, but it seems to occur consistently, regardless of which query is first in the batch.

Edit: There is a competing load going on, which is probably what threw off how the cost compares to the execution time due to the competition for CPU, RAM and I/O.