To elaborate, in my view there is a at least a two fold concern on this
particular ("Long term Mining incentives") front:

One is that the long-held argument is that security of the Bitcoin system
in the long term depends on fee income funding autonomous, anonymous,
decentralized miners profitably applying enough hash-power to make
reorganizations infeasible.

For fees to achieve this purpose, there seemingly must be an effective
scarcity of capacity.  The fact that verifying and transmitting
transactions has a cost isn't enough, because all the funds go to pay
that cost and none to the POW "artificial" cost; e.g., if verification
costs 1 then the market price for fees should converge to 1, and POW
cost will converge towards zero because they adapt to whatever is
being applied. Moreover, the transmission and verification costs can
be perfectly amortized by using large centralized pools (and efficient
differential block transmission like the "O(1)" idea) as you can verify
one time instead of N times, so to the extent that verification/bandwidth
is a non-negligible cost to miners at all, it's a strong pressure to
centralize.  You can understand this intuitively: think for example of
carbon credit cap-and-trade: the trade part doesn't work without an
actual cap; if everyone was born with a 1000 petaton carbon balance,
the market price for credits would be zero and the program couldn't hope
to share behavior. In the case of mining, we're trying to optimize the
social good of POW security. (But the analogy applies in other ways too:
increases to the chain side are largely an externality; miners enjoy the
benefits, everyone else takes the costs--either in reduced security or
higher node operating else.)

This area has been subject to a small amount of academic research
(e.g. http://papers.ssrn.com/sol3/papers.cfm?abstract_id=2400519). But
there is still much that is unclear.

The second is that when subsidy has fallen well below fees, the incentive
to move the blockchain forward goes away.  An optimal rational miner
would be best off forking off the current best block in order to capture
its fees, rather than moving the blockchain forward, until they hit
the maximum. That's where the "backlog" comment comes from, since when
there is a sufficient backlog it's better to go forward.  I'm not aware
of specific research into this subquestion; it's somewhat fuzzy because
of uncertainty about the security model. If we try to say that Bitcoin
should work even in the face of most miners being profit-maximizing
instead of altruistically-honest, we must assume the chain will not
more forward so long as a block isn't full.  In reality there is more
altruism than zero; there are public pressures; there is laziness, etc.