The majority of ratings seem to be warning people about red flags, not punishing provable scams. IMO this isn't a bad thing, since once someone has scammed, it's kind of too late.
What do you think about splitting the scam rating, with a "warning" rating for scammed previously OR you strongly believe that they will scam in the future, and a "scammer" rating for scammed previously AND you strongly believe that they will scam in the future? And then if you only have warning ratings, the indication displayed next to posts will be softer.
For both of these instances you've provided they deserve "Scammer" or the current warning, as it was a committed act. As techshare pointed out you do have the 3rd option already created, it just goes unused in a large amount of cases because it isn't as visible, and sometimes the feedback left is meant to hurt/punish more than warn others.
"Neutral" if displayed in the same fashion as positive and negative as a tally in the profile information. Ideally people could be persuaded to use this as their "Warning" or "Take Note" feedbacks regarding users. Encouraging the use of this system, a user who may have fallen into your example under "warning", could gt their red trust removed if they redeem themselves in the eyes of the whoever left the feedback.
Instead of having hard and fast rules we need people who can be flexible in the degree of feedback left to different situations, and apply the same reasoning consistently. I have noticed some newer DT members that do ask community advice on tags they've left or are considering leaving. Some are receptive to advice, some have made there minds up already and are looking for reassurance. Either way with this much further reaching DT network, we should ideally be able to get some diluted form of community consensus on most issues. Not everybody will be happy ever, but if we are at least willing to consider the other side of things we could see this system actually work. I think this can be done using minimal direction from the top; if people actively try and think about: Why they are leaving the rating? Is it supported? (reference) Is the information accurate How will this help the reader?