http://multivalent.sourceforge.net/download.html
If you run
java tool.pdf.Compress file.pdf
it will generate file-o.pdf and often trim quite a bit of weight.
That's probably what I would suggest sci-hub use too because it's easy to use and automatic.
EDIT: this looks quite promising too: https://github.com/pts/pdfsizeopt
But it has bugs with handling some images... converts them to black boxes. I'd only use this one if you manually verify the result first (so not a fully scripted process). (Or this could be due to me not installing all the bleeding edge dependencies from source as instructed.)
You could tweet at them with a link to a Github repo of your scripts, they are active on Twitter
Thanks for the tip about tweeting. I need to set up another github account disconnected from my professional identity but that shouldn't be too hard.
EDIT: and maybe a separate Twitter account. And maybe both accounts used only from public wifi, too. I wouldn't put much past the publishers when it comes to getting revenge on people who help sci-hub. Unlike Elbakyan, I am not out of their reach.
- Strip download watermarks ("Downloaded by Wisconsin State University xxx.xxx.xxx.xxx on January 12, 2017 13:45:12"). Many times, journals published by the same publisher do the watermarking similarly so you need write just one pdftk (or other PDF manipulation software) script for every journal under their banner. At worst, it's a one-script-per journal effort.
- PDF optimization. A lot of publishers produce un-optimized PDFs that could be 25% (or more) smaller with a completely lossless space optimization pass. This should save storage/network costs for access to individual papers and, more importantly, reduce the burden for bulk mirrors.
(I'd contribute the scripted passes myself if I had contacts within sci-hub.)