No this is not the intended behavior and as usual your keen eye has found a really bad performance issue with the current implementation. We need to note this one in JIRA and figure it out exactly.
I have a question regarding the objectClass index in the JDBM partiton. In JdbmStore the objecClass attribute is treated as system index. Hence any call of JdbmStore.hasUserIndexOn("objectClass") returns false. Hence the DefaultOptimizer doesn't annotate any objectClass filter but sets Long.MAX_VALUE. (I guess the same happens for entryUUID and entryCSN as they are also system indices). Is this intended behaviour?