Well, it's difficult to give a general explanation on that. The way the texts are stored in the game tend to differ per game. The most common method is to use a diskmonitor to find the Japanese texts and to replace them with English ones. Usually, however, the Japanese character set is not put straightly in the sectors, so even if you'd use a MSX with kanji ROM you may not be able to find the texts. You need to look for a way to 'decipher' hex values to Japanese characters.
When all that's done, the biggest problem comes: try to put your qualitatively good English text in the small space that is reserved for the Japanese text output. And there's the challenge.
As for me, reading and understanding Japanese games is no problem, neither is translating them to English, but this implementation process is quite difficult which is the reason why my translations take so much time before getting finished as well. I still wish there was someone living close to me who could do things like Adriano da Cunha did with Illusioncity - radically reprogramming the text routine so that it is perfectly adapted to English texts.
The biggest problem is translating graphical data (text that actually isn't text but part of a picture, e.g. the intro demo's of Firehawk, Undeadline, Rune Worth). Can't help you on that one, I'm afraid. For the games I translate, I get help from somebody else as well on this kind of technical stuff
