There are several techniques for timing jitter influence reduction are known. I’ve attached couple papers as examples, but you should be able to find more.
But, before “fighting with jitter” I would recommend you to test your scheme and confirm that that’s the main reason for degradation of your OSNR. As an example you can try to increase the span distance or bit-rate (as I mentioned before).
By the way, what was the increment of your Q-factor when you mentioned that your eye diagram stays the same while Q-factor is increasing?