I thought I had simple steps to reproduce here, but it is obviously not as simple as I thought.
Install multiple OU question types into a master install:
Run the Behat tests for --tags qtype_combined,qtype_ddimageortext
You will see two failures. The first one in qtype_combined is a known issue. The interesting one is the qtype_ddimageortext one. Now run that test alone: ..../question\type\ddimageortext\tests\behat\basic_test.feature. It passes.
So, something from the previous test is breaking this following test, which should not happen, but I can't work out what, and simpler examples do not break.
Obviously each scenario should be independent.