I think you're right. When I looked at block diagram of UFM it showed registers for both address and data. Since both serial and parallel access is available through megafunction I presumed that when in parallel access mode address and data is written/read as parallel to these registers. Now I think that megafunction converts it to serial data, clocks it into register serially and then uses it as parallel. Sounds like results that I'm getting are actually to be expected. That's not as exciting of a feature then as what I initially thought but still may be ok for some uses.
Thanks for your response, Mux!