I think we can assume that in the event of an absolute wipeout, UNS might give broadly accurate numbers overall but it would be useless for individual seats. In the event of a Labour landslide, there are seats where the Tories got 60% last time which are much more vulnerable than seats where they got 45%, simply because the Tory voters in the latter are much less likely to ever consider voting Labour than Tory voters in the former.
Yes, the Canadian example confirms this. In wipeouts like the PC's/NDP in 1993 and the Liberals/Bloc Quebecois in 2011, the survivors are a mix of ultra-safe seats, special local factors, and complete flukes.
Examples:
In 1993, one of only two Tory winners was in Saint John, a seat the Tories only won by 5% the previous election. Their candidate was the extremely popular mayor of Saint John and they actually increased their vote share slightly despite a 27% swing against them nationally.
In 2011, Bloc MP Maria Mourani managed to survive despite winning her seat by less than 1% in the previous election, thanks to a peculiar three way split.