The leveraged ETFs reset daily. In a day where the fund value goes up it needs to buy some more of the underlying (or contracts that represent it) in order to be leveraged correctly the next day. If fund value went down, it needs to sell some. This daily process - buying after price goes up, selling after price goes down - causes the time decay.
If you short a pair and rebalance every day, you will be doing exactly the same. The fund that went up in value increased your short position and you need to buy some more, and vice versa. The daily rebalancing will offset the gain from time decay.

In real life daily rebalancing will look a bit differently. The offset is not accurate. The tracking might not be accurate, and it may vary according to the time when it was made, or the execution. In a theoretical backtest one may see a small profit but this will most likely be 'eaten up' by the borrow fees, and will turn into a loss taking into account commissions, spreads, and slippage.

In order to make some gain with this strategy one has to take the risk of letting the position move away from market-neutral, allow drawdowns, and lose money if the market is directional with low volatility. In addition there are the risks of called shorts, possible imbalance in a flash crash etc.
There is no free money.

I made a backtest on shorting both TNA and TZA, with a theoretical daily market neutral rebalancing at the close. For the two years from close of 11/9/2009 to 11/10/2011 (without any trade friction) the strategy gained theoretical 19% on the amount shorted per each side of the pair (9.5% on the total shorted amount). I do not know what were the borrow rates over that period; today my broker rates are approximately 5.5% for TNA and 3.5% for TZA, a total of 9% /year for both sides. If these numbers are representative for the borrow rate over last two years, the compounded rate would have roughly taken away the 'gain' the backtest showed. This before commissions on the daily trades.