For parsing simple languages you can often get very far without ever using backtracking.

Actually, that's also the case for non simple languages. Perl usually isn't qualified as a simple language, yet parsing it doesn't require backtracking. Many languages, including Perl, use a grammar that requires one token to look ahead -- although sometimes perl cheats and scans (but not tokenizes) the stream ahead.

Limited lookahead parsing means that your grammar is written in such away that the compiler only needs to look at a fixed number of tokens before it can decide which grammar rule it has to apply to continue parsing.

If I were to make a language, be it an embedded one or a standalone, I would want the compiling phase reasonable fast. I certainly don't want any backtracking, so a full blown regexp would be out. I may use a bunch of simple regexes in the tokenizer, but the main parsing loop should be table driven or be expressable in a state machine. But no backtracking.