Yeah, those load times are insane. Most CMOD PDF Indexing is done on Windows -- simply because the libraries for PDF on the Windows platforms are the most heavily used, so get the most development time/effort.
There are other tricks, like disabling PDF's internal compression to let CMOD work its own magic under the covers.
You can test it out by running the loads manually from a PC with the CMOD server software installed locally. If the performance is way better, then consider adding that into your architecture.
-JD.