I was asked this recently, and my reply was

1. Make sure broker has enough CPU and does not swap
2. Check the LE storage statistics eg heappools and stacksizes and tune as recommendations using RTPSTG
3. It does little file system access, Ive never had to tune the file system.
4.  Some products use DB2 tables so check the buffer pool stats and that this performs well.
5. I remember something about caching the ODBC statements in DB2 - I cant remember the details but this is should be very common knowledge
6. Check MQ logs are striped (4 stripes) and on fast DASD
7. Make sure the MQ buffer pools no not get above 65% full.

For some message broker specific stuff

Increase the stack size 

Customising USS on z/OS