To put a bit more substance on the post below, I analyzed the riding-level numbers that Three Hundred and Eight calculates using his model. Under the scenario of 5% of voters leaving the Liberals for the NDP, applied uniformly accross all 308 ridings here's what I found:
The Liberals lose 18 seats they're currently projected to win (10 go CPC, 4 go NDP, 4 go BQ)
The NDP hold 2 seats the Conservatives are currently projected to win
The NDP takes 1 seat the BQ is projected to win
And the new seat totals would be:
I realize this is a very unrealistic scenario in that many voters are aware in their riding whether the NDP is competitive and probably would still vote strategically for the Liberal even if they decide they prefer the NDP, but the point was more to illustrate my claim about there being multiple tipping points. A 5 point gain for the NDP at cost to the Liberals would result in a safe Conservative majority, the Liberals would still be official opposition and the NDP would still be the fourth party, only managing to tie Ed Broadbent's historical top result (which was done in a smaller parliament).
In that result, could the NDP hold on to those voters? Many of them would regard this as simply a disaster and it would be hard to predict how loyal they would remain.
I think it is all or nothing in this game. The NDP have to overtake the Liberals and become the official opposition in order to solidify their gains. I don't know how they do that given that there are relatively few ridings outside the ones they're already winning where they are in second or a close enough third to actually win, absent a real tidal wave/landslide shift in support (like 10% instead of 5%).