My research group has gone pure python for coding experiments; we've been burned too many times by glitches and implicit behaviour in boxed experiment-building software to bother trusting it. Moving from a point-and-click experiment design interface to pure code does have a large learning curve, and you want to be careful to model your own code on well validated code from others (esp. for ensuring that you're implementing timing properly, which can be nuanced).
It may be tempting to hire CS students to code your experiments, but the danger there is they don't come to the table with the same experimental design background as you do and we've encountered some implementation errors as a consequence (ex. failing to check for input during dead time between stimulus presentation, etc).
While I recognize and indeed support the push to specialization in cognitive science, I do think that in the same way that we require all researchers to have a bit of background in statistics, we should also require all researchers to have a bit of background in coding, not least because it helps engender a mindset amenable to considering formal models of mind.