There could be multiple factors. Below are some i can think of
Queries which does lot of redistribution and results in product joins Queries consuming high cpu, IO, spool space usage (again this is relative as big table joins take high cpu compared to small table joins)
It's been a while, but if I remember correctly, impact CPU is a way of measuring the impact that a particular highly skewed query has on the system. It takes a look at the highest CPU across each AMP and multiplies it by the number of AMPs. This gives you a picture of the impact to system resources the query has.