Thanks, Ivan.
I'm currently running continuous logons and logoffs for two concurrent users with all 200+ Flex config files from the VMware Marketplace. The delay between logon and logoff differs for the two users, so the session "overlap" changes slightly for each run. So far I have not seen any delays after 100+ iterations.
Have done the same with both users concurrently performing a DirectFlex refresh in a tight loop; no delays seen yet after 400+ iterations.
My VMs are Windows 10 Version 1809, my file server is 2012R2, and I verified that SMB 3.02 is being used. I did not tweak any SMB settings, and left the caching configuration for both config and profile archive shares at the default "Only the files and programs that users specify are available offline".
ProcMon on the file server is showing lots of oplock-related entries, but no "OPLOCK BREAK IN PROGRESS". Just oplocks being requested, granted, not granted, closed, successful FSCTL_OPLOCK_BREAK_NOTIFY, etc. None of them seemingly causing a delay.
The repro attempt continues... 