I think this is a perennially interesting question because I wonder at what point do the odds begin to favor stock picking again. I.e., will the flight to active investing result in greater inefficiencies going forward?
While it is tempting to broadly declare that this is so, I suspect that market efficiency mechanisms behave like a Pareto distribution (i.e., 80% of efficiency comes from 20% of the effort). There was never a question about the market being efficient, but rather how efficient. So even if the flight to passive allocation results in better stock picking opportunities going forward, the residual amount of active allocation is probably sufficient to maintain the current status quo that it is hard to beat the market on a risk adjusted basis.
Moreover, I believe that the contemporaneous surge of "factor beta" (or whatever you want to call factor driven investment allocation) strategies is likely to supplant the traditional roles of active allocators, especially "closet indexers". Smart beta's dampening effect on inefficiency will become more pronounced as data sources and methods diffuse over time and as implementors grow in sophistication. As a result, fewer active allocators and dollars will be required to maintain the same level of market efficiency.
Anyway, I feel that it its rather pedantic to talk about the futility of active allocation on an abstract level. Broad platitudes about market efficiency are probably as inutile as are over-generalizations regarding entire asset classes' return expectiles.
To make any of these high level intuitions useful, one must apply them to allocation decisions. Given that baskets of securities and asset classes are far more likely to be efficiently priced than individual securities (i.e., it is far more difficult to identify inefficiencies), I believe that the greatest juice per squeeze will come from addressing "what degree of coverage, trading, or other arbitrage mechanisms are required in order for efficient price discovery of a given security?".
To my knowledge, this is still a very open question. Applying the 80-20 heuristic to individual securities suggests that only firms with very low relative amounts of coverage, liquidity, short interest, and correlations versus peers might be candidates for inefficient price discovery.
Another direction one could take this question is that persistent herd following behavior will result in tighter peer group co-movement. How will that modify the expectancies of performance outliers?