The situation you describe is not what you face in leetcode. The key here is "a problem space not seen before".
Bugs in production by definition are problems in a domain on which you are currently working and are familiar with. So yes in that case I would expect someone to work efficiently even under pressure.
But a problem in a completely new problem space no. Someone may know for example that a presented problem can be seen as graph traversal problem but may be blocked for the moment on the actual implementation of it. This is a context switching problem and not a performance problem.
