Over the years the discrete graphics adapter market has adopted a tiered system to sell as many SKUs per GPU as possible.
In general, although some OEMs and processes may differ slightly, here is how the third-party graphics adapter market works.
- A GPU designer/manufacturer (AMD/NVIDIA) designs a chip, tests it, and then mass produces it in an external silicon fabrication facility (usually Taiwan Semiconductor Manufacturing Company, or TSMC)
- The GPU maker will provide a "reference" design to each of its third-party board manufacturing partners such as XFX, Zotac, Diamond, Sapphire, EVGA, etc.
- Initially, the first products produced for each generation of GPU will be identical to each other as they are all just the fabricated reference designs provided to each manufacturer, perhaps with some minor component changes.
- At the same time, these board manufacturers are also working on their own customized versions that change various things about the boards... like:
- Chip binning - the process of selecting the chips with the tightest tolerances and a propensity to be able to perform better than reference specifications
- Passive/active component layout - Components like better voltage regulators, higher quality passives like capacitors, resistors, etc. are changed to either provide 'cleaner' power to the GPU (usually for the purposes of overclocking the part) or so that they can put that they did that on the packaging
- Custom cooling system - Usually the most obvious change between different SKUs, the reference designs usually only provide for a cooler that can adequately cool a GPU at its default specifications (and then not even sometimes, e.g. R9 290), this can afford the card to be cooled more and lower noise levels, useful for either overclocking or if you just don't want to hear a vacuum.
- There are a few other things that some board manufacturers do to alter their products, but you get the idea, they customize SKUs to fit different niches...
- A few months after the reference cards are launched, the custom SKUs start appearing on the market and based on how they are customized, economies of scale, market research, and hundreds of other boring business related factors, they are priced in tiers.
So to answer your first question, there usually is a benefit to purchasing an SKU that is not the reference design (read: lowest tier, probably lowest cost), as there are design changes applied that may improve the performance observed, longevity, and overall user experience.
To answer your second question, unless there is a massive overclock applied to the higher priced, but lower end part, OR the base performance of the lower end part was already comparable, no, the lower end part will generally not outperform even the lowest priced SKU of the next performance tier, usually due to the insurmountable architectural advantages like having significantly more execution units.
This is in general, there are variations on this structure by various brands, and yes some brands have better reputations than others for quality, but as an informed consumer you should probably read reviews of the products you are looking at done by reputable sources around the Internet. Hope this helped clear some things up.