I like the idea of faster and smaller kernel cycles. As you note the changesets/day ratio is actually higher than it has been, even though the changesets/release is lower.
having people waiting less time for the next merge cycle decreases the preasure to get a change into this merge window (even if it's not quite ready), and this tends to improve the quality of the changes, ....
remember that when Linus started the current development model, he was aiming at 2-month cycles
you should get the ancient kernel git repository and graft it on to the main repository and see what the longevity graph looks like with the additional data.