The trick here is that EV runs scheduled computations of the level of fragmentation in its databases and stores the results in the FragmentationResults table in each database (Directory, Fingerprint, and Vault Store).
When you refresh the monitoring alerts ("Re-run all checks" on the Status node in the VAC), it simply re-reads the results from the FragmentationResults tables. If the SQL fragmentation has been mitigated since the last time the computation was run, this change will not be visible to the monitoring system.
It is designed this way because computing SQL fragmentation stats can be an expensive database operation, and we don't want to bog down performance on the SQL server.
All that said, if you want to change the frequency with which this computation is performed, this article has the details.
And if you want to run the fragmentation computation as a one-off, that procedure is in this article.
Once the fragmentation has been mitigated on the SQL indexes and the recomputed into EV's FragmentationResults tables, only then will rerunning the checks on the Status node reflect the updated fragmenation state.
--Chris
