Your code fails with tests like this for instance.

Test.assertEquals(scramble('a','aaaaa'),false);

And because there are random tests, the number of tests your code fails is random too.

Paradox is that you raised an issue for a kata but it's your solution which is wrong.

If you checked errors separately and code produced desired result, but when tested in bunch it did not produce desired result, then you also probably should test the code with many inputs, and not separately.
If you still cannot find the problem with your code, you can post it here so we will take a look (remember about markdown formatting and spoiler flag when posting code!).

In javascript, I wrote various possible optimization solutions.
Errors where they should not be ... Every start programm in kata produces a different number of errors 10-30 (I checked the errors separately, the code produced the desired result)
What is the paradox?