Well, why would a hedge fund need a complex strategy. Think about moving billions of leveraged dollars around without trying to leave a foot print, or perhaps to rigorously explain a price pattern to your fellow phd peers that can come to your desk and tell that your work is crap. If my memory serves me correctly, I saw a huge multi-strat firm (Bridgewater) give a presentation where they described that type of flat meritocracy. The derivatives markets are obviously the new area for those that have superior math skills to make lots of money. From what I've heard, many of these complex strategies involve mean reversion (I'm guessing multi-vector index reversions) and perhaps inhouse factor models. If these funds are getting best of breed from IB's then why wouldn't they have models have some complexity to them. Then of course is the risk managment side, where they would have to hedge the "black swan" or other price shocks when risk-neutral strategies get killed.
As far as these quant firms getting killed, from what I have heard it's been from mean reversion models that don't mean revert and usually the developer is under pressure to compete in a flooded arena and they're trying to squeeze water out of sand. I think Andrew Lo wrote a paper demonstrating a simple mean reversion strategy that showed this phenomenon.
I'm not in this arena, this is just what I've heard working on the buy side through some networking and my own research. I guess one way to guage the trend of complexity is geneolize (probably not a word) known quantitative systems-- the turtles, demark, cointegration, donchian channels, etc.