If you have TN and your repo is getting huge, chances are that tn queries are the culprit. When you use the TNConsole or wm.tn query services to query transactions the results you see are paged. The unseen pages are continually loaded until the entire set is in the repository even if you only looked at the first page and moved on minutes ago. If this is your issue, constrain your queries more tightly. That is, instead of searching for all transactions, specify a date range at a minimum. For most development/debugging situations, TODAY or the last week is sufficient.
As for the repository holding this information in the first place, modern DBMSs should render this completely unnecessicary. DBs are great at paging and are a better place to do this operation. The data throughput waste, file system waste, and processing waste do not bring a worthy performance improvement with them.