Haakon84, my critique in post 103 still applies, but I'll be more blunt. Before you put any stock in these numbers, there are a couple of things that you need to do to fix them.
You need to fix the causation effect. Low save percentages can cause a playing to the score effect. See the data I posted on first period shots only. The range of save percentage is narrower than the range for the full game, because the playing to the score effect is not present to the same degree in the first period.
Why do you need to fix this? For goalies that play of bad teams, it's very likely that their low-shot games will come when they have allowed goals early, are behind, and the other team is playing defensively. For goalies that play on good teams, it's more likely that they are just facing few shots because of excellent defence. Goalies on strong teams are more likely to play with the lead, and it's easier to have a high save percentage when playing with the lead.
Second, you absolutely need to fix the selection bias effect - and I have no idea how you can do that. But it's a fact that when you select a group of games with low shots recorded, you are selecting 1) games in which few shots were actually taken, and 2) games in which too few shots were recorded. Underrecording of shots will cause save percentage to be lower. You must consider this effect before doing any adjusting!
Finally, why not look at seasonal numbers instead of game-by-game numbers? Seasonal numbers will eliminate the selection bias effect, as we would expect random variation in shot bias to basically cancel out after 50-60 games. We know the result is that we see no relationship between shots faced per game and save percentage. I think that is the best way of looking at this issue for now.