A long-only momentum portfolio is roughly equivalent to 50% long/short momentum and 50% benchmark, so the arguments still apply. A long-only strategy will reduce drawdown through diversification and potentially achieve higher risk adjusted returns, but you are still adding exposure to a skewed risk premium.
The most accurate way to backtest momentum would be to find a set of rules that were published many years ago and apply them exactly as written on market data following their discovery. Unfortunately, this will introduce survivorship bias since it is likely that unsuccessful rules have been forgotten. Similarly, looking at historical performance of funds which claim to have used this strategy is not robust. It is impossible to avoid survivorship and selection bias that far back because funds are so transient and information on the strategies they used is very sparse.
Any rule that has been tested on data gathered before its creation will be biased to some degree. Even if you avoid explicit in-sample curve fitting it is impossible for a person to ignore their own knowledge. Just the idea of testing momentum was not obvious 15 years ago.
Choosing to test dual momentum is a decision based on future knowledge (the fact that layered momentum strategies have worked) which implicitly biases returns upwards even if all testing is done on out-of-sample historical data. This doesn't mean that dual momentum does not work, just that the backtested returns must be biased to some degree. If the backtest was performed carefully the difference could be very small, but if parameters were fit in sample or rules were chosen using the investors knowledge of what has recently worked then the bias could be very large.
Men Faber showed that by simply lowering the shorted stocks by 20% for every 10% decline in the market, it effectively wiped out the previous momentum crashes from the backtest. Of course, you could also just apply momentum in a long-only basis and skip the crashes as well.
This is an example of trading rules that have been chosen to fit historical data . There may be compelling theoretical reasons as to why it works, but it is still difficult to put confidence in such a strategy going forward.
I believe that this is what Warren Buffett was getting at when he said that the first rule of investing was: don't lose money and that the second rule of investing was: see rule number one.
This skips the most important part of his philosophy, which is that risk is the permanent loss of capital. Both his words and actions show that he is not afraid of unrealized losses, and thinks that selling depressed assets is far riskier than buy and hold. (
http://www.mutualfundobserver.com/2013/10/permanent-loss-capital/ )
I'm not arguing against momentum as a strategy, I am even exposed to momentum in my personal accounts. I just find that most of the information available is not fairly represented - in particular the idea that backtests can represent unbiased future returns is dangerous.