The WF_STEP_INST table in the tririga schema basically holds the structure information for the actual workflow instances, i.e. all of the tasks and how they relate for all of the running workflows. This data will clean at runtime if you are in production mode. You shouldn't ever have too much data in this table as it should only contain the data for:
1) active workflows
2) workflows with pending tasks (e.g. user tasks)
Unless you have tons of workflows with user tasks that are leaving the workflows active (and hence the instance data in the system) or have a lot of orphaned data, this table shouldn't have too many rows. Removing the check next to Save Instance Data in workflow start tasks reduces the number of rows inserted (and eventually deleted) from this table. This can be important when doing large data loads, as I have seen delete operations against this table become a bottleneck when the system is processing large quantities of workflows after a data load.