It varies a lot by field. I'm a computational biologist and have dealt with both the CS and biology ways of doing things. In CS you send a formatted LaTeX manuscript to the publisher and they basically just print that, but the traditional biology way is just to send a Word document with little formatting at all. It is considered the job of the publisher to turn the Word document into something that actually is formatted.

Right (although this depends on the journal). But these problems are a function of Elsevier's immense market power as well.