What about lying? If enough miners claim to support larger blocks but actually don't, then part of the network will waste time producing blocks that won't be built on. IMO, if we want to put the power directly in miners hands it would be better to raise the limit entirely. However, to do so we would need to test the crap out of everything to be reasonably sure that there aren't bugs that are only uncovered by larger blocks like what happened when the soft limit was raised to 1MB.
I don't think it would be a problem. Like Erdogan said, the miners will use the "tip-toe" method of increasing the block size. Worst case, a large block gets orphaned and nobody tries again for a while. But if the larger block doesn't get orphaned, then the network will assume that that size is now supported (thereby setting a new effective upper limit).
IMO, if we want to put the power directly in miners hands it would be better to raise the limit entirely.
This doesn't put the power directly in the miners' hands. It keeps the power where it already is: in everybody's hands! It just makes it much easier for people to exercise the power they already possess.
However, to do so we would need to test the crap out of everything to be reasonably sure that there aren't bugs that are only uncovered by larger blocks like what happened when the soft limit was raised to 1MB.
I disagree. For example, I would not set my node's limit to anything greater than 32 MB until I understood the 33.5 MB message size limitation better. I expect many people would do the same thing. Rational miners won't dare to randomly publish a 100 MB block, because they'd be worried that it would be orphaned.
Furthermore, since miners would likely use the "tip-toe" method, the effective block size limit will grow only in very small increments, helping to reveal any potential limitations before they become problems.