Absolute no brainer imho, I'd take the real flying, and wish I'd been offered the opportunity to do so at that stage. After all if one was to model those options in a hierarchical way then raw data, unassisted flying would be the foundation for the flight directors then lastly autopilot/autothrottle. I'm under the impression that Sabena used to teach their students this way not only in the sim but that philosophy followed through into line training. Having flown with a few ex-Sabena guys now I've noticed their hand flying skills are excellent pretty much without exception.
The other factor to consider is for someone straight out of CPL/IR/MCC training, chances are their instrument scan is going to be $hit hot and so they'll be best placed to fly everything on the needles than at any other time in their career. That'll give them the best chance of developing a 300 knot brain if they're not thinking about how to fly.
Back to the real world now and the accountants tend to demand the lowest short term cost. Can't see it happening in today's environment unless you're fortunate enough to have someone that listens to and understands your opinion and also controls the purse strings. What slightly baffles me is that, arguably we're on the fringe of failure of our trusted automation centred teaching styles (ref various recent issues/articles re manual flying standards in the wider world), why isn't someone willing to give the alternative a shot, can it really be that much worse?