Your multi-document PDF has to have a repeating trigger string to identify the start of each PDF document. I don't think you can use the Generic Indexer for this purpose, since part of what it requires is the exact byte length of each document, and its starting byte within the container PDF. That may not be trivial to put in your .IND (generic index) file. For the Generic Indexer, in Windows, the two files share the same name except their file extensions.
The only Generic Indexer use I am directly familiar with is as a result of migrating data from the z/OS CMOD system to an AIX CMOD system. Each downloaded PDF, done by loadid, is broken up into all its individual PDFs. These we merge back together with a 3rd party PDF utility as one container PDF, and then loaded to AIX.
With CMOD version 8.4, the PDF Indexer on z/OS was too slow for practical purposes. All Adobe tools are optimized for Windows, and the PDF Indexer is no exception. The test PDF file indexed in a couple seconds from a Windows index server, loading to the mainframe CMOD instance.
I terminated the index job for the same PDF after two hours under z/OS. I then asked IBM about this, and IBM advised us at the time to index PDFs from Windows, which we did.
If you haven't, you should download redbook IBM CMOD Guide for good summary documentation on use of CMOD. It has a section specifically on PDFs.
http://www.redbooks.ibm.com/redbooks/pdfs/sg246915.pdf