The primary advantage of lower-cost reference cards is their potential bang-for-the-buck overclocking gains ("efficiency", some people would say).
A vendor like PNY makes only one (reference) GTX 980 card, no other GTX 980 cards. Many people speculate that your odds of winning the silicon lottery with a card like this from a manufacturer like this are increased because the manufacturer hasn't already culled out the best/fastest GPU ASICs (and GDDR5) to use in their higher-tiered (and more costly) products. They use every single GPU part they can get in the same product, and even if these are already binned by TSMC/NVidia, chances are that a number of better ones slipped through.
A second advantage with these reference cards is the NVTTM cooler. It was basically designed for a 250W card but is mounted on a 165W card. This provides a lot of spare overclocking headroom, even if the chips happen to be lemons. Non-reference "improved" air coolers are, ugly truth be told, sometimes inferior and often roughly equivalent to the (server-based) mighty NVTTM, only a few are actually superior - many card makers use their own custom coolers to market their brand identity and cut production costs.
My two EVGA-overclocked 1391MHz GTX 980 cards were CDN$875 each (after including backplates), and I am hardly able to run their clocks any higher than about 1450MHz - while I see a fair number of people clocking their "stupid" $625 reference cards at 1.5GHz and beyond. A price difference which could almost buy a third reference card in itself - and I strongly suspect 3-SLI at plain reference speeds (ignoring overclocks entirely) will still surpass 2-SLI at 15%-over-reference speeds in any real-world or benchmark metric.
"All opinions are not equal. Some are a very great deal more robust, sophisticated and well supported in logic and argument than others." - Douglas Adams
[/Korth]