You can fit FAR more signals on the bottom of a chip and in a multi-layer PCB than you can on any socket you might think of. There is a practical limit to how small the pitch of the pins on a socket can be if it's going to be field-replaceable; at high pin counts, the length of the socket itself becomes a driving factor.
As an example, even moving away from edge-connectors (which are just a non-starter for 800+ pins) Samtec's highest-density high-speed connector, the NovaRay line [0] advertises 224 signals per square inch, so an 800 pin module would require 4 square inches of connector space alone, not to mention making the system PCB assembly thicker and compromising on speed, power and thermals.
So an FM2 socket (to pick one of the more reliable ones), which has ~900 pins has a 4.5in^2 area and adds ~.25" in thickness, while reducing the maximum speed you can get through the chip at the very low voltages necessary to run LPDDR, which is what I meant by compromising speed and power.
Not to mention that socketting parts reduces your production yield over SMT, increases your assembly cost, and cuts reliability (it's fine on a desktop because you almost never move your desktop, and there's usually a big thermal assembly bolted on top).
Pentium m used to be socketed, pretty reliable, and available for thin laptops. I'm not trying to dismiss the difficulty, just saying they could perhaps be overcome
That's the thing, who is everyone? Such technology would only be relevant to premium-level ultraportable notebooks. Desktops, gaming laptops as well as budget laptops would stick with the cheaper and more ubiquitous DDR. Smartphones and tables will continue to solder RAM on due to space constraints. Apple (who is the driving force in this market segment) won't be interested since they are going fully custom and I doubt that an LGA socket will work with their on-package custom-build wide RAM interfaces. So in the end we are left with things like Dell XPS 13, MS Surface and few others. These are all popular and impactful brands, but they are just a drop in the bucket compared to the total PC sales. And I doubt that technology reserved to only selected premium laptops is going to be cheap.
If LPDDR is faster won’t desktops and gaming laptops also want it?
One possible other compromise could be soldered LPDDR RAM, but with a DIMM expansion slot for DDR as well. I wonder whether the DDR could be switched off and the baseline power draw eliminated if the RAM usage was below needing it.
So I will preface this with that I'm not an expert. But as I understand it, multicore systems are already frequently NUMA, but to a degree that most stuff (I am eliding categories of code here, but you get the idea) don't have to really care. From my understanding of how all this works, the difference of speed and bandwidth for LPDDR versus DDR seems like risking significant complexity for an expandability that few people would use and in turn cause an everything-or-nothing problem, where either high-performance code is written assuming anybody might have this, or nobody assumes it and you get a pile of stalls.
