Alright, I have some data now. I was hoping for a knockout punch and the ability to say “yes this has changed everything!” but reality is unfortunately a bit more nuanced.
I wanted to wait until I had data for a weeklong sale as well, since that is when the bulk of copies get sold.
CTR and visits
My methodology for this was to export the last year’s worth of impressions and click throughs to use as a baseline. It’s a lot of data so should smooth out any outliers, and I am also intentionally excluding launch data plus the period shortly after launch. It does however include the game’s launch on Switch which had a very positive effect on Steam sales.
I then exported data since the new capsule art was implemented. I took the impressions numbers from the old data and multiplied them by the CTR from the new data to create a number which represents “store page visits if the capsule art had been implemented a year ago”.
The result for this would have been a 15.2% increase 
in store page visits.
Sales result
Tricky one here. I’ve run multiple sales at 30% already, so it is likely I have already exhausted most people that are interested in buying at that level. On the other hand, doing a sale at a higher discount would be an unfair comparison as well. In the end, I decided to go tough on the art and had another sale at 30%.
I used two figures for comparison: The sales numbers for the most recent sale, and the trend of all sales at 30% (using Excel’s TREND() function). The trend includes the sale coinciding with the Switch launch, which is definitely an outlier.
The results for this were less encouraging, but not surprising. Like I mentioned, having had multiple sales at 30% already means I’ve “harvested” a lot of my wishlist base that wants to buy at that level.
The results were 16% decrease 
vs trend and 4% decrease vs the last sale. I’m not particularly discouraged by this, however.
Wishlist deletions
I was worried that using capsule art that doesn’t match game graphics might put people off. Here is a chart of daily wishlist deletions for the last 3 months. The red mark is where the new art was launched.
The two spikes are the frst days of weeklong sales. Users with the game on the wishlist get pinged about it, and some decide “you know what - I’m never buying this so I’ll just take it off my wishlist”. This is normal. To me, it looks like there is no change 
in wishlist deletions, which is a good thing.
Conclusion
Not sure if the cost of the art will be recouped by the effects on the sales at this point. However, it seems to me that it would have been well worth it to have had the art in place before launch. This is valuable information for future projects.