How can you say that, if you haven't spent significant time (with competent here, I assume someone who have spent year(s) in the relevant field) writing both assembler and C applications for the similar type of problem?

A macro may be good for copy-paste problems. But when you implement real algorithms, the regular lines of code suitable for copy-pasting, macros or similar normally represents a tiny fraction of the time needed to implement the algorithm.

The C compiler is not comparable to a strange macro assembler, where a C expression will map to a fixed code expansion.

And maintainance costs are representing a big part of the total costs for a product that is expected to live for a number of years and receive regular feature updates. Few embedded applications are just sequences of lookup tables with hardly any code in-between. And a number of studies have shown that the amount of work needed to understand source code depends on the amount of text needed to implement an algorithm. One C source line can map to many assembler lines, giving a higher code density - more work done in reloation to the effort needed to understand the code.