Optimizing work requires wrangling fairly complex, multi-faceted data structures like augmented graphs. Unless you do that in a very high level language in which that is as easy as possible, it's largely intractable for undergrad.

When I did this stuff in school, it was C on Unix. Nobody was going to get to a debugged data flow analysis of basic blocks with all the liveness info, optional register allocation, DAG-driven CSE, and whatever topics out of the Dragon Book. Not in one or even two semesters, with undergrad-level C skills, and a full load of other courses.

I think that working in Java wouldn't help much. You don't have premature free problem in Java, but in a batch compiler written in C, you can just avoid calling free.