I'm not honestly sure who I'd put my money on between the two of them. Neither is quite a prime physical specimen.
But yeah - the other issue Trump faces is that now he has multiple open lawsuits in which he's the defendant, so it's not out of the question that he could get indicted before November (though I think it's unlikely, he's every bit as much a wealthy member of the social establishment as Clinton and has a lot of experience at dodging lawsuits).
As to the poll numbers, ignore them if you wish, but in general ignoring the polls or assuming they can't be right is a pretty bad idea for a campaign. Sure, rural areas will mostly vote Trump bar maybe some bits of the west or northeast, but most Americans don't live in rural areas; 80.7% of the population of the USA was urban or suburban in the 2010 census, so a candidate could win literally every rural voter in the country and lose by a landslide. Polls tend to be a mix of online and phone, and then those numbers are rebalanced according to census data (so if you ended up interviewing say 45 urban and 5 rural voters in a sample of 50, you'd basically make the urban voters count for slightly less each and the rural voters count for more to make your data look more like the actual electorate). So there isn't really an argument to be made that the results are skewed by where or how they poll, and poll-based modelling has been very effective in the last few elections at predicting the result.
I guess the issue for Trump is that he's hit Hillary with the Benghazi and email things a lot, and it's still not broken through; it's hard to see where he gets the boost from to either really sink Clinton or boost popular support for his own campaign at this point. (This is, bluntly, speaking, not helped by the fact that he's running an unbelievably incompetent campaign and not exactly working hard to appeal to anyone who isn't already backing him).