Based on my investigations of the source code way back when, I think it has to do with the fact that a lot of the aligners are written in low-level languages and contain data structures and algorithms highly optimized for certain sizes of data. Sequencers that produce relatively long reads fall outside of the optimal range of some of these aligners. I'm sure someone will give a more elegant answer but that's a quick take.

You'll notice that even tools like bwa have run modes suited for different read lengths.