I understand that "back in the day" this was a very easy feature to program. Objects had fills and strokes, and that was about it. It was simple math to find the in between positions of the nodes and colors. Then objects started having gradients, transparency, special effects, and so on. How do you blend an object that has a radial gradient with one that has a linear gradient? You don't, it doesn't even make sense. So my suggestion is that objects need to have similar "attributes" to be blend candidates. I'm not a programmer, but if seems to me that if you limit blend candidates in this way, it would be an easy feature to add. Also, as I said in a response above, a gradient mesh would be a substitute in some situations.