FIR Cyclone2 M4K switch to Cyclone3 M9K No AUTO Block for storage

Is this a parallel FIR? Assuming each memory block feeds a tap of the filter, you're not going to use any fewer memory blocks when they're 9k than you would when they're 4k in size. In a fully parallel FIR, each memory block provides one coefficient or one data item each clock. Unless you do something special like running the memories at 2x the filter speed and sharing the outputs across two taps, you're going to be limited by the number of memory ports needed, not the amount of memory bits.