What matters to the bean counters is if it can be done without expensing something new, everything else be damned. Unless you can get a clear, popular government mandate to spend money to make things more efficient, this is not a palatable solution.

Given how large refactors tend to go in general, this also doesn't necessarily lead to good outcomes; even with a relatively technocratic administration led by Mike Bloomberg (relative to comparable mayors, at least), upgrading NYC's 911 system massively spun out of control: https://www.nydailynews.com/new-york/911-overhaul-2b-disaste...

Which would still face the same problem: the government would rather pay multiple times for crappy software at a lower price than one big bill for quality software the first time.

Or you can pay one big bill and still get crappy software.

It's difficult to evaluate the quality of software development contractors.

That's not a problem unique to governments, although the consequences in that realm tend to be worse.