I have had a total of 3 rMB's all the base 1.1/256. The first one was "OK". There was noticeable lag to the point of being annoying. The second one was atrocious, and is what I made videos of. This most recent one is buttery smooth by comparison to the first two. There is very slight studder, but what I see is the same as the Air/rMBP.
All 3 were fully set up with FV2 and all applications/processes fully completed. I have no idea why they would be so drastically difference. I can only think it has to do with quality control and possibly overheating on the first two. It almost felt like the originals were being throttled way too early and it was causing immense lag.
I will post a video later today of the new rMB just so you can see how different it actually is from the prior video. This one is silver and the other two were Space Gray, but I cannot see that being a factor.
You may be on to something - SoC / CPU die not seated properly on the logic board, leading to more aggressive thermal throttling behaviour.
Think about it on a full blown desktop - I've built my own gaming rigs before with i7 CPU's and custom coolers (air and water). If you don't get the CPU and cooling device seated 110% with just the right amount of thermal paste and positioning, you simply cannot try overclock or push the limits of your rig, and you will get instability and system OS crashes.
Same holds true for the Core M (and all core mobile parts) which essentially clock/turbo up and down all the time. I would imagine board contact and seating is even more critical and the tolerances even smaller in this application.