in reply to If at first you don't succeed ...

Just a thought, no fleshed out code: you could try to use diff, or its perl counterpart Algorithm::Diff or some derivate, as a basis to figure out if and where they're different. For each line where you spot a difference, you can check which value best approaches the value you're after.

This is under the assumption that the test results which be the same in structure, but may differ at some spots.