32-TQFP Arcing

I've been having some weird issues with my new PCBs. Odd (quiet) fizzing noises, failed communications with the processor, stalled processor... Generally really frustrating, and this might well be why.

Yesterday I noticed that the 5v regulator was getting quite hot, so today, as this is my first attempt at mounting something with a pitch as tight as TQFP I decided to start with a fresh board.
I populated just the minimum number of components necessary, programmed it to flash the 'safe' LED, and then left it running. It ran without issue for a while, and then I started to notice the fizzing sound again. This time I noticed these tiny arcs between the GND and VCC pins of the ATMega328p.

The soldering on the AVR isn't particularly clean - I've done a bit of rework to try to prevent the arcing, which has clearly failed. Could there be a slow build up of carbon until it starts to arc? and why have I never seen this on other boards? It also might be worth mentioning that there is a via behind these two pins, though I'd have said its further away that the neighboring pin...