I have experimented extensively with Lightroom and performance, but have no knowledge of its internal workings (and you would need Adobe to comment there).

It APPEARS that Lightroom has some sections of code that cannot execute in parallel. Over the years it has improved, but is not there yet. I have recently built a new system for photography which is all SSD, has separate controllers (one U.20 to get it off SATA), has 64G of memory. I tried to eliminate all bottlenecks, and it still tends to run about 50-75% busy building previews. In testing this I see no queue on disks at all, and certainly no memory pressures.

I believe you get slightly better utilization if you start multiple jobs on separate files, but it is not a full solution, and it does not seem to help to start 3, 4, etc. processes. Two seems to get all there is to get, and it is not much. My empirical impression is that certain activity is single threaded (probably by critical sections) and there is nothing the user can do.

Now that said, there is a lot you can do in general, for example: if your disks are busy, separating preview/ACR cache from images on separate disks (especially if spinning) and/or controllers, and in turn separate from catalog (a symbolic link is needed to separate preview cache from catalog), having adequate and fast memory (I have found it is difficult in single images (vs pano merges) to get LR to use over about 8-12G), choosing the smallest preview that you need ("auto" works pretty well), and finally by far the most important: using a CPU that has the fastest single code speed you can get.

If you use Intel that supports Hyperthreading, I find it performs slightly better with Hyperthreading off, not on (i.e. no "fake" cores).

A corollary of all this is that getting more, slower cores is not very helpful; getting few (say 4) faster cores helps much more. With significant code single threaded (apparently), as a benchmark to choose CPU's for LR, the single core processing seems the most important.

I do think 4 is better than 2; I certainly see at least three cores active at times. I do not think 6 would be better, much less 8 (this may change in later versions, and as mentioned is more about single core speed than number).

I also tried faster memory (clocking at 3000mhz vs 2400mhz) and found it helped slightly (about 10%). Faster GPU performance ONLY helps develop screen sliders, it will have no impact (as of the current version, 2015.10) with preview builds. Overclocking the CPU gives almost linear improvement in speed. Obviously overclocking anything may reduce stability.

For what it is worth, one often requested feature that is not in Lightroom that would help is the ability to use the embedded preview instead of buiding one. For many, with high volumes and low keeper rates (i.e. lots to cull), it is just too slow building previews to use. My personal solution for this was to stop using Lightroom for culling; I cull, crop and straighten outside with a faster tool that uses the embedded preview (Photo Mechanic in my case but there are many of them), and only take shots that have a high likelihood of being keepers into Lightroom.