The question is not "how to work around block devices with dynamically-assigned names at each boot", this can be done with UUIDs, labels or black magics.

At the time of 2.6.x kernels, we were all accustomed on that, once the physical machine was assembled, drives screwed in, net card inserted, etc, all that would remain exactly the same forever. At the first boot you had /dev/sda1, /dev/sdb1, and /dev/whatever, and that would remain the same forever.

Upgrading to linux 3.something trashes this de-facto rule. At each boot, you enter a new wonderland and even the root device may hide under a different device name, thus failing the boot unless UUIDs are used.

My question is: is there any rationale, advantage, or conscious tradeoff for why 3.x kernels trashed the straight, easy and effective device naming rules used in 2.6.x ?

I would rather say that was never so, but the algorithm used to detect devices found them usually in the same order. That was specified by how the detection algorithm was implemented. Using newer kernels you will get another, improved algorithm and probably also parallel execution (multi-threading) was introduced inside the kernel, so it gives more features, faster, but that non-documented and never-existed behaviour has been lost.

Correct me if I am wrong, but that is simply the method used for the default configuration? You can change it back to the old system you are familiar with so long as you are using physical volumes and not getting into more fancy LV stuff? That is to say, by default it is using these methods to better integrate into networked solutions such as hardware RAID arrays, fibre channels and the like.