Just like software optimization. When you've got '1000 things' to fix, you've really got 1 thing to fix: Process/Culture. There is a difference between premature optimization and not doing stupid slow things.
Even if you can eventually 'fix' the mess, it's now a mess with the inner loops optimized, not cleaned up. The original team is likely making another mess.
Not at all, you can be inefficient in non-critial code if it makes it easier to understand or maintain.
But if you're attacking a performance problem you don't start by wasting your time and money on little pointless fixes, you go for the root of the problem with extreme prejudice.
"Well we're still bottlenecked on the DB, but I made the pages load 5ms faster by spending the day optimizing our CSS."
And now nobody will understand it... thanks Bob...