I draw the opposite conclusion. If larger blocks have a higher probability of being orphaned, miners will make their blocks as small as possible, even significantly under the maxblocksize limit. This is why the debate seems so silly. We don't need a software-coded block size limit at all. blocks that are too big will get orphaned. This is happening NOW. It's why so few blocks are anywhere near the 1 MB limit.
I'd like to believe that myself, but...
That (the bolded part above) is true
now, but it might change. Think IBLT or other "block transmission compression schemes". Will there still be enough penalty in the form of orphan risk cost for big blocks?.
Makes me think of that proposal by Meni:
Elastic block cap with rollover penalties.