You don't want to run tests for all versions below 6.3?

this will have to be done cleaner/shorter, but also: why is this needed? any references for this?

Well, I looked into some of those examples file, and they will always fail (at lest with version 6.3), so why should we run them? And I am quite sure, that some of the tests were never run correctly with the previous versions (although I did not check them). For example, the (o)h3tr{3,4,5}.nw must run separately, and only (o)htr{1,2} must run sequentially. (Actually, the (o)h3tr2.nw is a restart of (o)h3tr1.nw. I could leave all the examples there but they will fail.

Honestly speaking, the examples in some case are outdated, and sometimes are not even implemented. Better way is exist to make the test, but the results should be always manually re-checked, beacuse we might have some unsignificant digits changed, or some numerical value orders are changed.

I can leave out the deletion part of the tests, but then it is not surprising that we have to set a limit for passed tests 50% or so.

Should I run all the tests then?

Good to hear you dove into these test, I was never able to (I lack the domain knowledge, for one).

Rather than removed these tests for a particular version of NWChem, just removed them all together from the list of tests being run, especially if you have serious doubt whether they ever worked at all.
Just add a comment with a list of names of tests that were omitted because they're known to be broken.
And feel free to reorder the dirdyvtst/oh3 and dirdyvtst/h3 tests to make sure they can actually succeed, no need to do that in a separate if clause either...