An interesting case study on the matters of scaling projects that use dynamic and interpreted language can be found in Beginning Scala by David Pollak.
I started searching for a way to express the code in my brain in a simpler, more direct way. I found Ruby and Rails. I felt liberated. Ruby allowed me to express concepts in far fewer lines of code. Rails was so much easier to use than Spring MVC, Hibernate, and the other “streamlined” Java web frameworks. With Ruby and Rails, I got to express a lot more of what was in my head in a shorter period of time. It was similar to the liberation I felt when I moved from C++ to Java...
As my Ruby and Rails projects grew beyond a few thousand lines of code and as I added team members to my projects, the challenges of dynamic languages became apparent.
We were spending more than half our coding time writing tests, and much of the productivity gains we saw were lost in test writing. Most of the tests would have been unnecessary in Java because most of them were geared toward making sure that we’d updated the callers when we refactored code by changing method names or parameter counts. Also, I found that working on teams where there were mind melds between two to four team members, things went well in Ruby, but as we tried to bring new members onto the team, the mental connections were hard to transmit to new team members.
I went looking for a new language and development environment. I was looking for a language that was as expressive as Ruby but as safe and high-performance as Java...
As you can see, major challenges in project scaling for author turned out to be in test development and knowledge transfer.
In particular, author goes into more details in explaining the differences in test writing between dynamically and statically typed languages in Chapter 7. In section "Poignantly Killing Bunnies: Dwemthy’s Stairs" author discusses Scala port of a particular Ruby example:
Why the Lucky Stiff... introduces some of Ruby’s metaprogramming concepts in Dwemthy’s Array in which a rabbit battles an array of creatures. N8han14 updated the example to work in Scala...
Compared to the Ruby code, the library parts of the Scala code were more complex. We had to do a lot of work to make sure our types were correct. We had to manually rewrite Creature’s properties in the DupMonster and the CreatureCons classes. This is more work than method_missing. We also had to do a fair amount of work to support immutability in our Creatures and Weapons.
On the other hand, the result was much more powerful than the Ruby version. If we had to write tests for our Ruby code to test what the Scala compiler assures us of, we’d need a lot more lines of code. For example, we can be sure that our Rabbit could not wield an Axe. To get this assurance in Ruby, we’d have to write a test that makes sure that invoking |^ on a Rabbit fails. Our Scala version ensures that only the Weapons defined for a given Creature can be used by that Creature, something that would require a lot of runtime reflection in Ruby...
Reading above can make one think that as projects grow even larger, test writing might become prohibitively cumbersome. This reasoning would be wrong, as evidenced by examples of successful very large projects mentioned in this very question ("Python is successfully used for... YouTube").
Thing is, scaling of the projects isn't really straightforward. Very large, long-living projects can "afford" different test development process, with production quality test suites, professional test dev teams and other heavyweight stuff.
Youtube test suites or Java Compatibility Kit sure live a different life than tests in a small tutorial project like Dwemthy’s Array.
Best Answer
The border between a fundamental type and an object is blurred and often artificially introduced. For example, In C a struct is just a bunch of records, just a derived non-object type. In C++, a struct is a class with all fields public, an object. Still, C++ is almost totally backwards compatible with C... the border is really soft here.
For prototype-based programming you need to have objects mutable at runtime. They MUST be soft-typed because each changes at runtime, a class of one kind changes into another - its type changes.
You might keep fundamental and derived non-object types as static though. But this introduces a weird disparity, objects are soft-typed, non-objects are static-typed, and a hard barier must be established between the two. Should you be able to morph a structure? A String? Should Number be a class or a fundamental type, or a set of fundamental types, int/float/bignum/etc?
It is just more natural and easy to learn, use and write to have this uniform, all types are mutable or no types are mutable at runtime. If you declare only one type (Object) is mutable, you end up with headaches and problems of both worlds.
Static-typed is:
Dynamic-typed is:
By blending the two, you sacrifice a lot.