Not 100% sure here but I believe the answer is yes and no. Yes to reading as this will round-robin bounce using MPIO iSCSI between the two nodes. No to writing as, with synchronous replication, the two nodes have to be kept identical so a write operation goes something like this:
BEGIN
Write to node #1 via iSCSI MPIO
Replicate to node #2 within StarWind over replication network
Acknowledge the write
END
So there is actually a performance overhead with writes and HA which is why you need to ensure the replication network is as fast as possible (10GbE point to point) and dedicated.
However, with StarWind v8, the performance benefits of the new logging file system plus secondary caching will offset this.
Even if you have a big write-back cache in front of the spinning tin, the same replication has to be carried out from the two caches. Consider the situation where block #565689 is written to node #1 and this goes into the write-back cache. If another iSCSI initiator tried to read the same block from node #2 before replication, it would get bad data.
I suppose you could code around this so that node #2 is somehow informed that block #565689 is stale but this still requires communication across the network and it might be faster to just transfer the actual block.
But I'm guessing
![Wink ;-)](./images/smilies/icon_wink.gif)
Although I did think as you did at first before StarWind mentioned the above.
Cheers, Rob.