All compression techniques depend on image content, of course. The more detail and noise the less the compression. Thus, "YMMV" is accurate.
But I can't find any image of mine that has so little detail that a 16 bit TIFF/LZW version is smaller than a version with no compression. Makes me curious about the two images whose sizes you show here. Are they typical, average photographs? Or are they test images of a blank sky, shot at ISO 50 and out of focus?
I selected an image from my Canon 5D3. Average scene detail, ISO 400 with some noise reduction applied in ACR. The following snippet from Bridge shows the file sizes of several 16 bit versions. The base "no compression" version is 126.71 MB. The LZW version is 148.95 MB. That's 22.24 MB or about 18% larger.
A version with a heavy dose of Topaz Denoise brings it down to 139.85 MB. Still larger than the no compression version. So I created 3 more versions by filling the image with increasing amounts of blank gray. As expected, the LZW compression becomes more efficient as more detail is removed. When I finally destroy enough of the image, the LZW version becomes smaller than the no compression version.
In a previous test, I processed 100 images from a landscape session. I made two 16 bit versions of each, one with no compression and one with LZW. All the LZW versions were bigger, by 10% to 20%. Some of those images contained 2/3 sky with little detail. Some were exposure brackets for HDR and contained significant blown out or clipped portions (zero detail).
https://kellyphoto.smugmug.com/Other/Temp-Upload/n-5RJTN/i-BtmzQRv/0/O/i-BtmzQRv.jpgHere is a link to the image I used, if you or anyone else wants to test with it.
Would you be willing to share the images you used to show LZW compression is efficient?
https://www.dropbox.com/s/kqyqy6yp0pjeyj7/08%20Original.dng?dl=0