Reality vs expectations (2.7x vs 2.8x)

It’s been a while we’re on 2.8x now so I was wondering if there are plans to fix/improve the following;
1.Apart from most official scenes (bmw, classroom etc), most of my scenes do not see any noticeable speed-up while renderimg on GPU, some of them are a tad slower too.
2.Scene building (cpu-bound) seems to be optimized for 1 or 4 threads most of the time for most scenes, with small bursts @ 100% all-core cpu load. It’s ok for single renders but it is quite a drag for videos, where all those seconds you loose per frame add up each time.
Have tested various cpus/core counts and frequencies to confirm this and also excluded memory bandwidth as a problem; the gains are proportional to cpu frequency and IPC but after 4 cores it just looks like the remaining cores are mostly on vacation while the scene is being set up before rendering. (24 core cpu vs 4 core builds just 25% faster or less).
Also checked 2.82beta because there was mention of build speedups in the feature log but it resulted slower actually.

Thanks in advance to anyone that might help!