What I think is happening is:
- timer works according to expectations - you can observe that by looking at number of step 1 node executions
- every trigger of business rule task register listeners to be notified on rule execution
- in case rule throws an exception it is not successfully executed meaning that listener is not signaled
- as soon as the other process instance is started and goes into business rule task that is successfully executed with trigger all event listeners as they are matched based on ruleflow group name
So depending on number of business rule task instances that were created with failing rule, you should see that number + 1 triggers of step 2 node.
IMHO you should never thrown exception from the rule as that causes not predictable results.
HTH