The query plans are indeed different. The actual plan on TEST has a nested loop, an index scan and a key lookup. The actual plan on BISQL parallelism and a table scan. So the reason for the longer elapsed time with the shorter cpu time might be I/O? They both produce the same amount of records