The ink limit settings at the top of the descriptor file is for the initial calculation of the overlap points of the other lighter gray partitions, and the default ink limit is only used if you do not define ink limits for individual channels (I prefer to leave the default ink limit as 0 and then define each ink limit on a channel by channel basis). The ink limit and cross over settings are just the starting point for roughing in a partitioned profile, and then refining is done with the gamma, overlap, highlight/shadow settings.
If you look at .quad file in a text editor, it is a % of the 16bit value (0-65535), and the ink limit is just a percentage of that 65535 output value. The numbers at each of those 256 steps is what controls how much ink is being spit out per channel. The original image pixel data isn't being changed, but the amount of ink being deposited does.
There is enough overlap built into the QTR curves creation algorithm to hide these dots as the ink load ramps up to what is perceived as full coverage. When you examine a print of the ink separation page to determine the ink limits you are not only watching for where the density stops increasing, but also watching for where there is full coverage before excess bleed. From what I understand, the way the ink is laid onto the papers—not how much, but which nozzle is firing and when—is controlled by the dithering algorithms in the underlying gutenprint driver/program.
A reasonably low ink limit is also a great question. When I was testing uncoated papers, the limits needed to be very low (low 20s) because the dots would bleed into each other and start clumping, giving a muddled appearance. With those low ink limits I could also start to see printer dots from the darker dilutions being dithered too heavily, but I couldn't compensate with more overlap from other shades because of the clumping problem. From what I have done, with 3rd party 4 to 6 partion ink sets, I would say that, in general, the lowest I would set ink limits would be in the 30s. For Espon K3 inks, it is hard to say, and dependent on the paper. I don't have any ink separation pages handy from a k3 printer but i keep them in the 50s-60s for the black gray inks and in the teens for the color toning inks.
While you can change the gray curve as you mentioned to achieve a similar result as changing the ink limit it is also bypassing some of the other controls in the descriptor file. For making prints I prefer to leave the gray curve blank (or "0;0 100;100") and use the correct overlap, and gamma settings to get close to linearity, and reserve the gray curve option for slight tweaking of the final linearized profile.
Hope that gets close to answering your question...
Richard Boutwell