A couple of thoughts.

1. Your entire 30,000 companies by 12,000 products classes and 500 groups can be stored in a flat file in 45 MB.

   This can be slurped into memory or scanned record by record. Either way, the results of your 350 queries can be determined in ~100 seconds.

   The data is stored 1-bit per yes/no answer. 12,500/8 * 30,000 == 44.7 MB.

   The code to generate a test DB, generate 350 test queries, and run them, is 50 lines. (No formatting of the report is done.)

2. If this is survey data, it is static once the survey is finished. So why does it matter how long it takes to generate the report?

   You only need run the queries once and store the results for reuse each time the web page is needed. Or better still, generate and store the html.

   If you want the viewer to be able to sub-select the results produced for them, then generate a separate html snippet for each of the 350 categories and merge the required bits once you know what they want.