I use MetaTrader 4 for backtesting.
From my experience I suggest to use "Every tick" testing model, always use date period, check Recalculate checkbox - after date range or currency change (you may always keep it ON), download M1 and all timeframes history and test only on history period which has M1, all this will make backtesting more close to reality.
And do not try to run optimization on big periods, test on month or weeks, then change period, otherwise testing will take too long and sometimes strategy tester generates it's own bugs or just freezes.
I believe that Metaqoutes will make tester in MT better in next version, in general I like this platform very much.