I put FvM's "unsigned_multiply_accumulate" unchanged into my design, added an additional register to the output datapath (to be sure to have no influence of the subsequent datapath elements), and let it run. Same result: The adder (accumulator) is outside the DSP block.
I noticed that the signal path goes directly from a_reg/b_reg to adder_out, but besides a longer path, I'd not expect influence on the mapping (no registering of multiplication output, although the signal is called 'mult_reg').
Is there any other setting I might have missed that prevents integrating the adder into the DSP block?