You can usually trust Benchmark (I don't know about Benchmark::Timer)... but it's not perfect either: Benchmark.pm: Does subroutine testing order bias results?

Sometimes, timing the full program (or a script specially written for the ocasion) from the shell with time is also a good solution.