It sounds like you have just curve fitted your MA crosses to your historical data. From your ohther posts it seems that you seem to favour faster time frames so I am guessing that the actual amount of historical data in your 100k of bars was actually not very much actual real time to base your curve fitting/optimization on.

The trouble with MA’s is that their behaviour is very different for trending markets or sideways markets so with no filter to identify what market they are in they will either perform great or terrible.

The other thing to consider is what other possible curve fitted elements are in your strategy. Fixed take profit and stop loss levels could be the real curve fit culprits as it is generally easier to curve fit an exit than it is to curve fit an entry.