It depends. You save on compile time after the first run (which is relatively small for most queries), but the majority of variation in CPU/IO time is more likely to be due to whether the pages to satisfy the query are in the buffer pool or not, which will depend on the amount of memory, the frequency that query is executed, other workloads on the system and the runtime parameters.

Elapsed time can obviously be much more variable based on other load on the server, concurrency,locking etc.