Lemme see... All numbers assume completely random choices on the part of all players, not reasonable, but the only way I can think to model it, I'm also assuming an initial GGVB distribution, that the vig isn't one-shot, and that the two townies are completely vanilla:
No-lynch/no-vig:
66% -
Assuming equiprobable lynches the next day, there's a 33% of win, 33% of lose, and 33% of VB going into night. How is this usually handled? Do they both end up dead? I'm going to call it a scum win.
33% - GG
Assuming equiprobable lynches the next day, there's a 33% of win and 66% of lose.
So, unless vig vs. scum isn't an outright scum win, the vig doesn't actually do us any good if we no lynch today, and we have an expected win of 33%.
No-lynch/vig-kill is slightly better, because it gives the vig a 33% chance of pulling out a win tonight, plus the 11% of the time that the vig and scum kill the same player, you've still got situation A above, for another 33% win. Aggregate likelihood of winning, ~37%
Lynch:
25% GG
50%
25% GGV
So, the aggregate with a lynch today followed by a vig kill tonight is 66%, which clearly towers over the other two options. Of course, I don't have any good information on which to base a choice.