The problem is that if you're writing code that you want to be as fast as possible, then you're going to want to pin down the size of those integers to be as small as possible in order to let the compiler take advantage of the vector units found in modern processors. For example, if you're writing a lot of graphics code, not only are you going to be wasting space if you're giving every color channel of every variable 64 bits (8 bytes) of space, but you're going to be giving up a lot of performance, because the compiler has a lot fewer hints when it tries to autovectorize your code. Or, as another example, if you're writing a low-level networking library, then things like byte-level alignment matter. Reducing the integral types may make it simpler, but makes the language itself a lot less useful.