Recorded pronunciations do apply to commands as well as dictation.
Though, the process is intended more as a way to enable the recognition of words that aren't in the dictionary by default (E.G. brand names); Improved recognition because the recorded pronunciation more closely matches what you're saying, rather than what the default language rules assume the pronunciation should be, is more of a side effect, in my opinion.
It's important to mention that speech recognition in general is sensitive to accents, E.G. if you don't speak with a US accent, the English - US recognizer will be less reliable.
For general recognition issues, the '"VoiceAttack is listening, but what it hears is complete gibberish compared to what I said"' section of
this topic also offers some suggestions.