How valuable was this paradigm shift?
What a silly question. In dollars? No one can measure that, can they?
The impact, however, is pervasive, so there must be considerable value in using OO instead of procedural.
Exactly how necessary is OOP for the development of large, complex and expensive software?
"Exactly"? What a silly word to use.
Clearly, large, complex and expensive software is built with non-OO languages. COBOL, VB and plain-old C are used heavily today to built large, complex and expensive software. So it's not exactly necessary at all.
How feasible would it be to write Microsoft Visual Studio, Wolfram Mathematica or a CAD software in an imperative-procedural programming language? Has it been done before?
What a silly question. The Mathematica kernel is written in C with some "OO extensions". Essentially C. Autocad is written in C.
What is the effective size and complexity limit for software written in a non-object oriented language?
What a silly question. There are no limits.
Let's think for a moment.
Pick an OO language at random.
Python. It's written in C. So you can always build an OO framework in a procedural language, then use the OO framework to build your application. You've used a non-object oriented language, and yet, you have the benefits of object-orientation.
Java. It's written in C.
C++. It's written in C.
All programming languages which are Turing-complete are equivalent.
The limits are mostly patience.
What are the most impressive (large and complex) and important (highly used) softwares of this kind?
"this kind"? What kind? Procedural code?
Linux. Apache. MySQL. Python. All written in C.