Most real-time development can and is done using a C compiler. Again speed is only one issue for using ASM, certain levels of kernel mode device driver code are in assembler because of the Intel microprocessor callgate or privilege level transfers.

The key to fast code in whatever language, interpreted or compiled is a basic understanding of how your code is going to operate. Function call stack frame overhead, loops, case statement etc. all add overhead whether interpreted or compiled. Look at how easy it is to create an O(n2) algorithm in *any* language, David Frankenbach gave a fantastic explanation of this in message 843573.

Sometimes ASM is best because its all you've got.