I think it is next to impossible to measure the performance of the garbage collector without knowing the details of the implementation. Different implementations may show very different behavior given the state, and the exact state is hard to determine without at least knowing the architecture of the memory subsystem.
Measuring the GC using just an Applet is of course non-sense, it doesn't normally have access to a stopwatch. It would only work with a combination of applet and PC side framework.
Generally you should try and avoid GC at all times. Only request it or release references when absolutely required, preferably do not use it "in the field". Hopefully that tip will alleviate the need for GC performance comparisons.