Journal papers are single-blind, i.e. the reviewers know the names of the authors and can make decisions based on reputation, so if you publish great (e.g. faked) results once, it becomes progressively easier to do it again.
Reviewing for conference papers is typically double-blind - we don't know the identity of the authors, and although we can try to guess, a couple of studies show we're wrong most of the time. Most papers (65-85% at good conferences) get rejected, so it's easy to flush something that looks fishy. (then again, it's hard to tell the difference between someone playing games with their numbers and someone who just has the typical crappy CS systems understanding of statistics) The addition of artifact evaluation to a lot of conferences provides another avenue to flush out faked results, too.
However conference reviewing is vulnerable to reviewing fraud, since reviewers and authors typically come from the same relatively small community - see https://news.ycombinator.com/item?id=23459978, or just google for the words "computer science reviewing cabal".
Reviewing fraud might not be quite as destructive in the long term, as the papers that get published through such fraud are probably more crappy than faked, and will hopefully not get cited. In addition, when reviewing fraud is uncovered it's much more of an open-and-shut case - a single person can fake data and make it look a lot like good data, but review fraud requires communication between people who are explicitly barred from communicating on that topic.
A lot of it depends on the community, and the size of the community. When you have 1000+ reviewers for a conference things will always get out of hand - I know of someone who got his cat registered as a reviewer for a particular SIG conference that he hated, just to make a point. I assume one could get a lot of fraudulent research published in that particular conference.