I think that extreme outliers, particularly if they are sponsored titles, should be removed or trimmed to the nearest in-line data unless there is a logical reason.
That should of course be done in both directions.
Although tbh, I don‘t think doing so would change the overall results in a noticeable way.
The term outlier doesn't really apply well in this context.
If a developer makes the effort to optimize a game for Nvidia or AMD (not both) rather than releasing a game that's not optimized for either platform, the latter developer doesn't deserve to be rewarded with greater praise. It's better to be half-optimized than not optimized at all.
The main problem is when games are intentionally sandbagging performance by doing things like tessellating invisible water. That doesn't just involve trying to harm a competitor (e.g. AMD's cards which weren't as performant in tessellation at the time). It can also involve the mindshare gimmick I mentioned in my prior post. Simply releasing a 'demanding' title is often enough to get a lot of press focus. Many consumers assume that a 'demanding' title is better, more complex and therefore more worthy of interest/purchase.
DOOM was an 'outlier' because it was optimized so well for AMD hardware. Does that mean reviewers should have eliminated it from roundups? Quite the opposite. DOOM showed how a customized engine could extract a lot of performance being left on the table by others, when running on AMD.
Another interesting example is Deserts of Kharak, which actually ran extremely well on AMD's Piledriver CPUs, CPUs that trailed Intel badly in basically every other game. As an 'outlier' it was actually more worthy of journalistic focus because it suggested that there was performance being left on the table, that more optimization may have helped Piledriver performance. That doesn't mean it should have been done but it is very interesting for tech enthusiasts to learn.
Optimization doesn't have to involve sandbagging one platform. It is possible to optimize for both, switching the code to suit the platform being used by each gamer. It takes more effort. Many companies coast along on extremely inefficient design (like Civilization) because they aren't challenged by competition and by having journalists hold them accountable for their laziness. It's also very arguable, in the case of a game like Civilization, that the inefficiency of the language/coding style used is fine because it enables more content to be released. That's the basic fight between low-level optimizers and high-level content spewers. It's hardly an easy argument to resolve. Many times there is a middle road taken.
The market currently appears to favor churning out content quickly rather than producing polished optimized products. 'Eternal beta testing' as the consumer experience has been a concern for many years now. Proponents of that, though, argue that consumers benefit by getting more content. Getting more less-reliable content is better than getting less content that's more reliable? Many consumers seem to have been indoctrinated into living in the conditional future, rather than demanding that the present and near past be more satisfactory. They're always looking for the promise of great products and quickly abandon recent ones for their flaws. How much of that is indoctrination and how much of it is innate temperament I can't say.