Look again and re-assess. There is no exaggeration.

Both tests set and reference the variable 1000 times each. The reason for including a loop inside the both benchmark subs is to amortise the cost of that subroutine call across the 1000 iterations of the code under test. Without it, the code would be benchmarking the calling of the benchmark sub, not the code under test.

If the benchmark is not to your taste, then consider this. Setting and then referencing a single scalar value involves:

• Using the object
  1. 4 hash lookups
  2. two subroutine calls
  3. two multi-element list assignments
  4. one conditional test.
  5. plus the cost of stacking parameters and returning the results.

  And finally one each of setting or reading the value of the actual scalar.

• Using a variable
  • one each of setting or reading the value of the actual scalar
  • Errrrrr. That's all folks"

All in all, it is quite remarkable that the pointless OO code is only 7 1/2 times slower.