I'm afraid differences in resources are partly induced by Sql Server caching (intermediate) results between the first and latter queries.

If you use DBCC DROPCLEANBUFFERS between the queries, caches are emptied and differences in time and physical reads may better reflect actual use of resources.