flok3r:
Sorry if that wasn't clear enough.
Its not a source problem for sure. The current draw of the test installation is below the rated 50mA for the 3.3v pin of an UNO, but my iPod (used as source in my project, providing regulated 3.3v) provides even more.
The 3.3V supply from an Arduino is (for some boards) rated at 50mA absolute max. On connecting the load it will try to draw much more current briefly to charge the decoupling capacitors, which could cause problems?
The standard configuration of a 328 for Arduino is with brown-out threshold set to 2.7V, which is a little too close to 3.3V for comfort - perhaps the brownout detector is kicking in before the 3.3V supply has stabillised?
It is essential that AVcc and Vcc are within 0.3V of each other, BTW, someone's already hinted at this.
If you run at 3.3V you must ensure that no pin goes above 3.6V or below -0.3V at any time.