Surprisingly, the 6280 can keep up with the 68k on quite a few 16bit stuffs. Sometimes coming very closet to matching it, other times beating it. But like stated before, comparing snippets of code between these two processors isn't exactly representative; even simple addition comparisons.
But here's one:
(68k)
move.w d0,abs.w ;12
add.w d0,abs.w ;16
28 cycles
(6280)
lda <$01 ;4
clc ;2
adc <$02 ;4
sta <$01 ;4
lda <$03 ;4
adc <$04 ;4
sta <$03 ;4. 24+2 = 26
That's just a memory+memory->memory operation; 16bit+16bit->16bit. The 6280 beats it by 2 cycles. I used ZP on the 6280, but I also used the MD's fast access mode ram (top half of the 32k) for faster accessing too. Not exactly the same, but fairly close in principal (if you treat certain ZP variables as ram instead of data regs). If you have a matrix of variables that need to be added to a single value (say like moving a series of linked objects), the 6280 automatically gets free indexing and such. I had quite a few examples of 16bit operations, and even when the 6280 was slower cycle wise - it was still pretty close. And of course if the operation is 8bit+16bit->16bit instead of a full 16bit operation, the 6280 gets a nice little speed up there (a branch to skip on non overflow detection). This isn't to take away from the 68k's 16bit operationally ability (it can do quite a few fancy 16bit things), it's just to show that different between one being 8bit and the other being 16bit isn't as drastic as it seems.
Of course if you design the game logic specifically around the 6280 strengths, it'll scream along no problem. The key is high level optimization in design along with low level optimizations. This might limit you in game design slightly, but the player/gamer will never know. The 68k is more flexible and robust overall, while the 6280 is extremely fast at more specific/narrow-focused designs.