Failure in the way PeteM mentions, i.e. a capacitor case swelling (indicating it's being stressed), in 5 months compared to 16 years is stretching the "sh*t happens over time" scenario a bit I think. I suppose some of the questions the designer must ask are how's the component expected to fail and what are the consequences of it failing? after they've done all the sums.
In this case when the parallel capacitor fails O/C or low capacitance. i.e. high impedance, the series ones certainly would be stressed even more(even if there was still sufficient starting torque). Eventually something would go pop I reckon.
Consideration should be given to how long the component's likely to last based on factoring in reliability based on prior experience or the component manufacturers data and the conditions the component's operated under. If it's stressed in any way then it's likely to fail much sooner rather than later. Again it's up to a designer to work this out.
Does it have the potential to fail with with a pop, bang or fizzle after a few months or just go low capacitance and/or the motor fail to startup with no other consequences after 16 years? Personally I leave this to the designer who should carry the can for poor design.
The whole point of changing like for like is that you avoid adversely influence the reliability of the device you're working on and/or how the replacement component's going to fail, by avoiding making the wrong choices of what alternatives to replace with, based on limited skill or information.
