Does the degradation in performance closely follow the spike in IOPS on the SQL Server?
I can't call it degradation. It's just the VMFS datastores latency which goes up. But it does correlate with the SQL server IOPS load. And I do see the issues with the cache. We have frequent back-to-back CPs, but again the reason for it is simply the load from the SQL server.
The solution to this problem was to defragment the SQL server volumes. They became fragmented because of the small amount of space inside the aggregate. Now disk busy time dropped from 70% avg to 40% avg.