During debugging of the ClassPools, I discovered that, for every transformation, the same classes were being retrieved from the classpool again and again during pointcut matching, all invocations being indirect invocations originated in org.jboss.aop.pointcut.Util.
I have locally added a very simple cache (that definetly will need improvement if it proves to have a positive impact) to Util class. This cache basically stores the result of previous queries, which not only skips calls to the ClassPool, but also avoids recalculation of the entire query result.
As soon as I'm able to run the tests without failures, I'll measure the performance gain we may have with this cache, if any. All my findings will be posted here.