Any chance you have links to before/after tests? Make sure you run enough tests to get a consistent result (3+) but then looking at the waterfall will tell for sure, particularly if the images are not in the critical path.
Remember that download time is not linearly related to file size. What you are interested in is the number of tcp round trips - which will, of course, depend where the requests are in the tcp slow start cycle.
As a for instance, I have a 120kB background image which last time I measured it came down in the following numbers of packets on a new connection:
3, 5, 8, 12, 21, 37
i.e. 6 round trips. To get it down to 5 round trips i.e. only a 1/6 improvement, I would have to reduce the size by 37/88.