L2s - what are the bottlenecks constraining performance?

There's always attention on gas fees and the improvements that upgrades like Dencun will bring.

I don't hear much said about the factors constraining throughput and finality. For instance, Eclipse looks very interesting, and the online docs make the case that such an L2 with parallel processing will combine high throughput while benefiting from the security of Eth, but I don't see much mention of the bottlenecks of posting data to Eth.

How performant would you expect it to be?