Most visitors will not hang around for the page to load -- I thought that the demo was broken when I first visited the site. A few ideas:
JSON is not a compact data format as the tag names get repeated in every datum. CSV/TSV is much better in that respect as the headers only appear once, at the top of the file.
On the other hand, repetitive data compresses well, so you could set up your server to compress your JSON data (e.g. using mod_deflate on Apache or compression on nginx ) and serve it as a gzipped file that will be decompressed by the user's browser. You can experiment to see what combination of file formats and compression works best.
Do the summary stats need to be calculated every single time the page loads? When working with huge datasets in the past, summary data was generated by a daily cron job so users didn't have to wait for the queries to be performed. From user feedback, and my own experience as a user, summary stats are only of passing interest, and you are likely to lose more users by making them wait for an interface to load than you are through not providing summary stats or sending stats that are very slightly out of date.
Depending on how your interface / app is structured, it might also make sense to split your massive file into segments for each category / material type, and load the categories on demand, rather than making the user wait for the whole lot to download.
There are numerous other ways to improve the load time and (perceived) performance of the page -- e.g. bundle up your CSS and your JS files and serve them each as a single file; consider using image sprites to reduce the number of separate requests that the page makes; serve your resources compressed wherever possible; move the JS loading out of the document head and to the foot of the HTML page so it isn't blocking the page contents from loading; lazy-load JS libraries as required; etc., etc.
