We have had very good results using diversity with madwifi drivers on OpenWRT. Generally we will use one antenna with vpol and one with hpol. Driver will sample SNR off both antennas over a time period and then switch to use the best antenna based on sampling. Most effective in high motion/dynamic environments. Can also help with space diversity between two vpol antennas where multipath is changing (over water, train yards, etc).
It would be great if MT supported this, but you have to pick your battles.