The setting maxDataSize has nothing to do with the retention of old data, but only with the maximum size of a hot bucket. (therefore the rotation of hot to warm)
Make sure that another setting was not sneakily deployed, like frozenTimePeriodInSecs, or maxTotalDataSize, or any volumes limits, or even repFactor if it requires more copies to be made and reduce your actual shared storage space for that index.
To investigate :
- run a "splunk cmd btool indexes list" on your indexers to check
- and look on your splunkd.log for any bucket freezing events on the indexers, they mention the reasons why the buckets were deleted/frozen (size or time constrains)