Clock for clock comparison says Broadwell already gets up to 5% faster. Then, with the thermal headroom gained with the process shrink, one can imagine current Haswell 15W i5 at 1.4/2.7 GHz as found in the base MBA11 to get a bump at something like 1.6/2.8 MHz and being able to not throttle that fast once turboboost kicks in. We'd be closer to a 15-20% effective gain in comparison to what we have now in the low-end at equivalent power; there is benefits with a die shrink.
Of course, there probably will also be lower-end 15W Broadwell chips that would fit in and give only a marginal gain over the previous generation in performances (but then with thermal and power consumption going down).
Perhaps that is. There are benefits with a die shrink, but I am not sure how much it will benefit in terms of performance and battery life.
You can't really expect much better than Haswell IRIS performances (rMBP13) going down to the 15W Broadwell chips, and Haswell IRIS Pro performances (rMBP15) now available to 28W Broadwell chips as should be used for the next rMBP13 revision.
So if the 12" retina Macbook gets a 2560x1600 display, graphic performances wouldn't be much better than on a current rMBP13.
That's at least why we could try to make sense of the rumored lower retina resolution 2304x1440. But looking farther than Broadwell, that really would look like a bad idea; it looks good only at 2x, any other non-integer retina factor are nice to have but shouldn't serve the default usage, and 1152x720 is a downgrade compared to what we have now.
The rumored 2304x1440 only makes sense from two perspectives, at least for me. First, it offers the same ppi as 2560x1600 on a 13" MacBook Pro, so it seems like this resolution didn't just come out of nothing. Second, it would allow a better battery life than a 2880x1800 screen, and battery life is critical for the next MacBook Air, given that Apple is not supposed to take a step back.
----------
[edit: wow and that thread reached 1k post!]
... and counting.