They may be designed to "last" 20 km of tunnelling - or whatever the number may be, depending on type - but that doesn't mean that they don't need maintenance. Main bearings only last a quarter of that distance or even less. Hydraulic lines fail, cylinders give up the ghost, conveyors wear thin, bearings need to be replaced, the list goes on. They are machines that operate in extremely harsh conditions, and their cost can not be equated to just their purchase. The risk with a used machine is that it will not be as reliable as a new one, and with that comes the potential for increased costs - be it in terms of repairs, or in terms of delays.
Newer machines are faster, and more automated. What required a crew of 50 people 20 years ago only needs 25 today. That number may be reduced further into the future. Speed increases aren't as substantial, but according to people in the field machines today are about 10% or 15% faster than those 20 years ago.
I frankly fail to see how you can equate the cost of a $36mil machine - which is designed to be disposable - with the vast ongoing costs of everything else involved in building a subway. $100mil or $400mil per kilometer, either way the foregoing of the cost of one brand new TBM doesn't save you the cost of one kilometer. There are far better places to save real dollars on a project.
Dan