I'm recently working on improving the scanning algorithms and the entire scanning workflow.
The improvements focus both on robustness and precision, so they could make your results better - if you would like to try them, you can download the unstable version of Voxelizer (help -> change release type -> unstable).
I'm still working on one issue which causes the point clouds from the left and the right laser not to always be perfectly aligned - as soon as this is dealt with, a new stable version will be released. I also hope to prepare a longer tutorial on the entire scanning process.
Light and models
As for the light, what is important is for it not to be too bright, so that the lasers are clearly visible. I scan daily with a scanner on the floor beside me, without any special setup and the scans turn out just fine.
Your models seem scannable - the most important thing is to have point samples that cover all surfaces of the models. The new version features a partial scanning mode, making it possible to only scan a small angle of the model that wasn't caught by the full scan.
The suggested workflow is now to make 1-2 full scans (with quick scan, now the default scanning mode), and then, if needed, to capture missing surfaces parts with the partial scanning mode.
As for meshing parameters, the two most important ones are interpolation weight and samples per node, so I would try modifying those as a beginning, starting from interpolation weight as 0 and samples per node as 5.
Interpolation weight forces the reconstructed surface to 'stick' more to the scanned points, making the reconstruction more precise. The presence of noise in the point cloud, however, may cause the surface to be more jagged, which is why I would start with 0.
Samples per node forces multiple points to be taken for each surface fragment, so it's also a noise/precision tradeoff.
Additional explanations, albeit more technical, may be found on the website of the authors of the reconstruction method, here.