Well, RB insinuates pretty directly in http://www.mameworld.info/ubbthreads/sho...part=1&vc=1 that the actual development compiles take significantly longer -- at least longer enough that it's cutting into actual work time without getting a major upgrade.
Remember, this project is on a volunteer basis. Things that are going to make the codemonkeys require newer hardware for a non-paid project are likely to be shelved. Especially in this economy, demanding things that are going to be serious inconveniences to the devs aren't going to be taken with less than an entire bucket of salt.
That'd probably be the biggest reason right now. It's simply too much of a pain at the moment. Nobody ever said it's going to be like this forever, mind. There may be a complete change of attitude on it in another three years.
Or putting it down to the absolute simplest equation:
*THIS* is a bad time to force such a changeover.
While I actually SUPPORT the idea of flat-out merging the two as a single release, I believe that it's a long term goal at best and not really a great idea for the moment.
Try checking the MAME manual at http://docs.mamedev.org