That correlation (in wins G% rises and A% drops) also works with such guys like:
View attachment 1020794
Now that I've had time to review the data, I can say that this argument is false. It looks like you carefully selected a few players to "prove" your desired outcome. (What made me suspicious was the inexplicable exclusion of Lemieux).
I looked at the breakdown for the 100 highest scoring players in NHL history (coincidentally, these are exactly 100 players who have scored 1,000 points). I used the same method - games were classified into either "wins" or "everything else" (which includes regulation losses, overtime losses, shootout losses, and ties). I then calculated the split between goals and assists for each player.
Example - Daniel Sedin won 670 games, in which he scored 268 goals and 441 assists (37.8% of his points were goals). He played in 636 games with any other outcome, in which he scored 125 goals and 207 assists (for essentially the same breakdown - 37.7%).
Looking at the top 100 players together, the result is very similar. In wins, goals comprise 39.5% of these players' points, and all other outcomes, goals comprise 38.8% of their points. (In practice, this means that if a player scored 100 points over the course of the season, and his team won and lost 40 games each, we would expect the player to have something like a 20 goal / 30 assist split in wins versus a 19 goal / 31 assist split in losses).
All the names you presented rank very high on the list in terms of skewing towards goals in wins. (I don't dispute that some players follow this pattern - but many of them don't, and you didn't acknowledge that reality). There are a lot of big names that have very similar splits (call it +/-2% in either direction - including Lemieux, Crosby, and Jagr). There are several players who do the opposite, and skew heavily towards assists in wins (Bourque, Clarke, Esposito, Sakic, Yzerman, Bossy, etc).
We all enjoy reading about new ideas here, but you need to make sure you're giving an honest summary of the data, and not just cherry-picking the examples that support your argument.