What I've been noticing lately, is that lowest and highest prices occurr when the EMA diff. percentage stops growing (or decreasing).
Looking at the bot's popup, I noticed how the lowest prices often have the lowest ema difference percentage, as the highest prices occurr when ema-diff is inverting its trend.
With the current strategy, the bot's behaviour would be this: (graph w/actual bot data, interval is 30m, thresholds in grey +/- 0.25%, first graph is ema-diff, second is price)
http://imageshack.us/a/img835/3959/senzanomeqg.png
So, the bot would have bought around 119 (friday 17 @ 13.30 utc+2) and sold around 120 (yesterday, sunday 19 @ 19.00 utc+2), with a price difference of 0.8% (diff=1$), just enough to cover Mtgox fees.

This is because of the ema indicator's inertia: once price has begun to decrease, ema stays positive for another X samples (X varies greatly), and while the SELL threshold is reached, price has lowered a lot.

If we search for price maximums and minimums in the bot's popup, one may notice how price peaks occur at the same time of EMA-diff peaks. In other words, we should need another indicator for monitoring ema-diff, not the price! (Like two emas based on the ema-diff as the signal, not the price. Like we need the second derivative, not the first one.)

Look at this:
http://imageshack.us/a/img521/3978/senzanome2y.png

Let's concentrate on points 3 and 6 for now, ignore those yellow circles. So, once we've established point #3 is a maximum of ema-diff, it should be a good time to sell: price is likely going down for the next samples (the bot would have bought around 113/115, check for yourself as the bot doesn't return older data). Similarly, buying @point #6 would be a good choice.

Now: since at present time we don't know shit about the future trend of ema-diff, how do we know we're in a good peak?
First of all, we're not in a peak if the following ema-diff is greater than the current one. (So a trading delay of minimum 1 sample is intrinsic)
Second, peaks around the 0% line (like 5, 7, 8 ) are to be ignored.

So, it seems like accepting as maximums/minimums those points
1) above a certain threshold
2) followed by a trend-inverting sample
is a good solution.

How about peaks 1 & 2? They have to be ignored! Peak 4 (which is the actual price maximum btw) is going to be ignored? (should be!)
Any help is appreciated, thank you.