Already, the Obama approval markets have included both the RCP average and also a separate specific pollster market, Gallup in that case.
An analogous pair of markets could be a good idea for Clinton vs Trump as well, both RCP and non-RCP.
The current Clinton vs Trump polling market is interesting, but, in practice, predicting the RCP average is largely about guessing which pollster comes out at which time.
A fuller explanation and suggestion:
Presently, any signal of actual rise or fall is easily drowned out in the noise, as can be seen from looking at the history on RCP.
As an example, in the 3 available past cases of comparison, NBC/WSJ polls each have had Clinton 8% to 11% more ahead of Trump than in Rasmussen polls coming out soon before or afterwards. Other information I have read on those respective pollsters suggests that pattern is probably not just a temporary coincidence, and in fact it can be expected to repeat in the future.
The existence of the RCP poll market is fine, but maybe we could also have a market or two based on specific pollsters? The latter would end up being more like once a month markets, but that could be okay
I'd suggest maybe two markets:
1) The current weekly RCP poll average market
2) A semi-monthly PPP market
Obviously, the exact pollster choice(s) for Clinton vs Trump are disputable. However, PPP is a pollster which many of us have seen in a multitude of state primaries beforehand, probably one of the most reputable and recognizable other than Gallup. Even the best pollsters exceed theoretical statistical errors in predicting elections, if unable to fully take into account differences in passion and turnout, but, overall, PPP seemed to do decently aside from that.
Optionally also, personally I think Rasmussen might be a possible choice for an additional (third) market, having a slant from its automated polling or weighting but apparently a relatively consistent slant not prone to wild fluctuations, with little merely random volatility claimed by them in advertising and seeming to fit history.
The scientific terminology for Rasmussen performance might be only moderate accuracy but high precision, with a systematic error but better such than large truly random errors. However, that is debatable.
If not having 3 markets (such as RCP, PPP, and Rasmussen), just having 2 markets (such as RCP and PPP) would still be an improvement.