Qsan NAS products use inline deduplication. All data being written are checked and deduplicated before actually written. (Post processing means the data is firstly written and later on deduplicated.) 
All data stored in the zfs prior to enabling deduplication cannot be further deduplicated.
Storing the DDT could be very memory-consuming. It is therefore suggested to always have SSD read cache disks configured for using deduplication feature.
The dedup limit displayed in the dashboard merely calculates how much deduplicated data the system can afford to store in the system RAM cache without SSD cache enabled, and displays how much memory quota (allocated for deduplication usage) has been used.
If the deduplication limit (for system) is exceeded:
1. If the dedupped zfs resides in a pool with SSD read cache, it's not affected.
2. If the dedupped zfs resides in a pool without SSD read cache (where the DDT in this case is stored in the system RAM cache), the dedup feature for this zfs will be automatically disabled and cannot be enabled again. All later-stored data will be stored with standard non-dedupped manner and therefore cause the zfs to become a hybrid instant with both dedupped and non-dedupped data inside.
ZFS metadata structure works in a much more complicated manner than that of other file systems such as ext4. Modifying any part of the data structure involves actions of updating relative checksums and metadata all the way up to the root layer and the ZFS uberblocks. This becomes most obvious when the operation containts lots of write/modify/delete actions to large amount of small files.
With mutiple copy (more than 1) configured for a zfs, the write performance will be reduced by more than 50% (two copies) or more than 67% (3 copies). No obvious impact on read performance can be observed. 
The "number of copies" options offers the possibility of keeping up to two more "shadow copies" of data stored in every single block. Using 3 copies in this option would therefore occupy 3 times capacity for storing the same amount of user data. 
No, thin-provisioning only has effect on space allocation of zfs. Enabling thin-provisioning or not does not have any impact on peformance.